Introduction To Artificial Intelligence Technique: Humans."
Introduction To Artificial Intelligence Technique: Humans."
hy~c, oplimiution, geotechoical and transportation is artificial neuron, that is; a simple mathematical fi1g. 9.2.l: Example olllmple utlftdal neural network
solved not only in need of mathematics, physics, and
mcclianics calculations but also depend on the
C08111Caing.
model (function). _ In order to have the benefits of mathematical
experience of pnctitiooen. 9 Such a model has three simple sets of rules: complexity that can be attained through interconnection
•1.1 Some deffnltlona of Al . d activation. At the
This .knowledge and experience are illogically multiplication, summation an . led of individual artificial newoos and not just making
• Wlnaton
incomplete and impreciae, and they canoot be bandied entrance of artificial neuron the inputs are weigh . system complex and wunaoageablc, usually artificial
by traditional procedures. However, artificial - "AI is the · what means that every input value 11 · ul · lied with neurons cannot be iotercoMected randomly.
thin IIUdy of Ideas which enable computers to do m bp
intelligence has its own superiority. It can solve gs which make people seem intelligent." indi~idual weight.
~ ConatnJction Mana ement SPPU)
Introduction to Artlflclal lntell ence T8Chn
lier M eonetructlon Manapement (SPPU)
- There are severa) "atandardir.ed" topographies of
artificial neural networb. These ~fined
9-3
Hidden layers are layers betWeen the input and oUIJ>ut
layers. The neW'Ons in ~h layer are uauauy full~
connected to the neurons m another layer. The three.
-- Toe input units are directly COllnected with the output
unit anatid they · are genenJly uaed for pattan
9-4 lnlloduction to Artlllclal lntellgenee Techn~ue r
.., Wortllng
- The featurea of lhe biological neural netwoat are defineS the activation state of the network at that the network.
Dendrites receive the input through the synapses of
attributed to ill IIIUcnn and function. instanL Likewise, the set of the N output values of the _ The process of adjusting the weighl8 is refemd to u
Olher neurons. The soma proceases these incoming
netWork definea the output state of the networlc at that learning. Once the teaming process Is completed, the
- The fundamental unit of the network ia called a IICUIOll signa)J over lime and converts that prooesaed value into
or • nerve cell. Fig. 9.6.lshows a schematic of lhe an output. which i1 sent out to other neurons through instant.'° final set of weight values corresponds to the long term
llrUclure of a neuron. lhe axon and the synapses. - Depending on the discrete or continuous nature of the memory function of the network.
_
Introduction to Ar1lflclal lntellgence Techn~ue r
As an outcome; in many CIICI it ia capable of aolving
- ::.Uon of vaguely similar patterns while PIOVidin
- probabilistic ~ g , and genetic algorithms. the problem in appropriate way. Even milling input
mean, to deal scientifically with subjcctiVity g
- The oewaJ nctworb exhibit mapping capabilities, lhat die . h 'all •1 Combination of techruques from all theac areas are information has even now turned out to be aatiafactory
i1, they can map input pa11em1 to their mociatcd tenitOJY lhat traditional science u eaacnli y iSllortd_
UJCd to design an intelligence system. in knowledge-based syatema.
output pattern,, In other words, fuzzy. l°".c is concerned ~th Plllting Neural networks · provide algorithms for learning, - Puzzy logic pcmrill to lower complexity by allowing
- The neural networb 1ecn by examplea. Th111, neural away from logic that 1s cnsp or Boolean (binary; 0 Or the use of imperfect data in SC11Jible way. It can be
classification, and optimization, whereas fuzzy logic
network architectura can be 'lraincd' wilh known I). implemented in hardware, aoftware, or a combination
dea)s with issues such aa fonning impreasiona and
examplca of a problem before Ibey are tested for their of both. In other words, fuzzy logic approach to
Thia melhod has an advantage over Boolean logic io reasoning on a semantic or lingui1tic level.
'inference• capability on unknown U1J1aDCea of die - problema' control mimics how a pmon would inlJce
problem. They can, therefore, identify new objecll 11111 it mi.mica complex bwnan reasoning so aa to lltive Probabilistic reasoning deals with uncertainty. decisions, only much faster.
previously Ulllrained. 11 realistic conclusiona about the imprecise and often Although there are significant areas of overlap between The fuzzy logic analysis and control methods shown in I
- The neural nctworb poucu the capability to
fuzzy niture of reality.
neural networks, fuzzy logic, and probabilistic Fig. 9.10.1 can be described u : •
gcocnliz.e, Th111, Ibey can jedict new OUIComea from - The past few yem have observed a rapid &rowth in the reasoning. In general they are complementary rather • Receiving one or large number of measurements
pultmida. number and variation in applicationa of fuzzy IOgic than competitive.
1 or other asseumcnt of conditions existing in some
- The neural nctworb are robuat syslems and me fault (FL). Fu1.zy logic methods have been uaed in image- Recently, several intelligent systems called neuro-fuzzy ayatem that will be analysed or controlled.
tolmnL They can, thcref<n, recall fulJ pa11an1 from Ulldtntanding applicationa such u detection of cdg~
incomplete, partial or noisy )llllans. systemB have been used. There are many methods to Processing all received inputs according to human
feature extraction, clwilication, and clustering. combine neural networks and fuzzy logic techniques based, fuzzy "if-then" rules, which can be
- The neural networb can )IIOCCl8 infOllllllion in
- Fu1.zy logic p01ca die capability to mimic the hllll]an however, it is essential to understand basic ideas in the expreased in simple language words, and
paRIJcl, at high lpeed, and in a dillribured IIIIIWr,
mind to effectively employ means of l'eaSoning that art design of fuzzy logic techniques. combined with traditional IIOll•fuzzy procasing.
9.9 Fuzzy Logic approximate rather than exact. In traditional hard 1
Characteristics of Fuzzy Logic Averaging and. weighting the results from all the
computing, decisions or actions are based on Jncisioo, 9.9.1
. . (SPPU • May 12)
individual rules into one aingle output decision or
certainty, and vigor.
• Introduction The casential characteristics of fuzzy logic are related signal which decides what to do or tella a
- Precision and certainty carry a coat In soft computing, to the following: controlled system what to do. The result output
- Fuzzy Logic WU initialed in 1965 by Lolfi A. 2.adeh, tolmnce and impreaaion are explored in decision In fuzzy logic, exact reasoning is viewed as a limiting signal is a pm:isc defuzzllied value.
profcuor for computer ICiencc • the Univmity of making.
California in Bakeley. cue of approximate reasoning. The following is Fuzzy Logic Control/Analysis Method
- The exploration of the tolerance for imprecision and In fuzzy logic, everything is a matter of degree. diagram.
- Basically, Fuzzy l.ogjc is a multivalued logic, which uncertainty IIDdediea the reIDarkable hwnan ability to
allow, intcmiediatc valuea to be defined between In fuzzy logic, knowledge is inle!pretcd a collection of
uodentand dilto!ted speech, decipher sloppy elastic or, equivalenlly, fuzzy constraint on a collection
conventional ellaluati0111 like true/false, vea/no
handwriting. comprdiend DIWICca Of natural
high/low, etc. Notions like radiec tall or vay fl8;can ~ of variables.
language, SIIDIDlariu text, and RCOgniz.e and clllsify
formulated DWhcrn•tically 111d !lroceaed by imagea. Inference· is viewed as a process of propagation of
80 18
cotnputm, to apply • IDOl'e human like way of elastic constraints.
thinking in the PIOgrlmming of COIDpUlcn, - Wilh Fuuy logic, mapping rules can be specifying in
Any logical system can be fuzzified.
- Puzzy logic is a comp1cx 11111bema1icaJ method lbat IClma of Words rather than nwnbers. Computing wilh
the w<Xds cxplorea Ullpreciaion and tolerance. In Fuzzy Fl&,9.lo.t
~owa aolving difficult limuJatcd problems with many 9.10 Fuzzy Logic System
1Dpull and output Variablca. logic, another basic concept is the fuzzy if-then rule. _ With the purpose of operating fuzzy logic needs to
- :'11hough rule-baaed systems have a long hiatOJy of use Today control systems are usually described by be represented by numbcn or descriptions. For
- Fuzzy logic iJ capable to give l'elUIII in the form of mathematical models that follow the laws of physics,
in artificial intelligence, but machinery for dealing with example, speed can be represented by value 5 mis
~ ~ o n for a 1pccifjc inlerval of output llltc, stochastic models or models which have developed or by namtivc "slow".
IO II II 0DCeUary Chat tbi1 lllllbcmatical method . fuzzy COlllequen11 or fuzzy &Dlccedents is missing in
luch aystems. from mathematical logic. Tenn "slow" can h•vc diverse meaning if used ~y
llrictly ~ from the more familiar I · u
_
A general trouble in such constructed model is ho~ to different pcnons and must be understood with
~
loch 18 Boolean algd,ra. llglCI, -
1PPlications, -
moat of the an fuzzy logic solution
move from a given problem to a proper mathematical respect to the observed environment
- It i1 fiequcnuy lllggCaltd that the pow« of lhc h ~~ lranafonnation of a human solution. Also, fuzzy 'nl
modcl . CcrtaJ y,
today's advanced computer
. h Some values arc easy to categorize, while othen
brain iJ • flmction of ill ability to efficienu wnan ogic can IIIOdcl nonlinear fwictiona of uul'trary
A ...
0.8
,,
,,,
I
I \
\
\
I
\
I
·.
Fls,9.10.2 a
0.4
, I
I
The shape of the membership function should be 0 I \
0 20 40 eo 80 100 I
I \
\
9.11 Fuzzlftcatlon illustrllive of the variable. However this shape is also 0.2 __ , , ,I \
I
I
~-,
.. 1. Trlanplar MF Fls, 9.11.3 \
value ia compatible to I memberahip function, 0.4
\
\
\
value from 0 to 1, also known II truth value CJC Specifie3 by two parameters {m,o) as follows: '
fuzzyinpuL
M-benblp llmctloa, MF: Definea a fuzzy 11C1
I 1
c,0.8 Gauasian (x : m, o)= ex"l -
J (x-m>2}
'i,2'· • .
0.2
0
-10 -6 0 5
,_
10
by mapping cri1p values from ita domain to the Q. (b) 'c'
fO,e Where m and o denote the center ~ ~idth of the
IICla IIIOCiated degree of membenhip.
jo.4 function, respectivelY, A small o will generate a "thin"
Crllp Input■ : Dillinct or exact inputa to a certain lo.2 MF, while a big o will lead to a ''flat" MF. o.a
1y11em variable, usually mcaa!Wd paramct.en
external from the control •)'Item. 00 .. 4. o.e
Label : Deacripave name uacd to identify a
membcnhip function.
.U·70JJMF 0.4
0.2
I
Scope : Or domain, the width of the membenhip Specified by three par&meten {a,b,c I as follows: O.I
0
o., -\0 -5 0 0 10
function, tho range of concepea, usually nwnben 0.2
over which a membership function i1 llllppCd. ' oo ao _, eo ., too (d) Cbangln1 '•' and 'b'
➔ L
lntl0ductlon to Al1lllclal lm:'!P:':: Ttchnlque r
Ac:ce.lblllt7
populllion CTellel a new population. Thia new· Exert ayatema 1R1 alway1 available for 111C when human
population la llled to generate aublequent populallona ,uch 81 biological arma l'IICCI, hoat-paruite co- expert■ are not readily available.
and 10 on, yielding 10luti0111 that arc cloaer to the 1. Optimization evolutlona, symbloel1 and reaource owe in ecologlea. .. IL Comllteacy
• optimum aolutlon. -t 7. population pnetia modell
2. Automatic programming Expert 1yatems are lea■ likely to contain inacancica
_. 11 'The valuea of the objective lunction of lhe lndividual1. OAJ h81 been uaed to atudy queation1 In population provided lhe expert 1yatem bu good knowJedp
< of the new populllion an, apln detennlned by 3. Machine and robot IMmlng genetlc1, auch 81 "under what condition■ will a gene for ~talion In IICCUl'lciea or erron can be eully
recombination be evolutionarily viable?' Interactions prevented
decodlna the atrinp. The8e valuea expreaa the fitneu
4. eoonomlc ~ between evolution and learning: OAI bu been Uled to
of the aolutlona of the new gencrali0111. ➔ UL 11me coaatnlnta
atudY how individual learning and 1pecica evolution
- Thia complete, one cycle of genetic algorithm called • 5. Immune ayattm modlll The numben of copica or an expert 1yatem can be
affect one another. . made wbereu, the training prnceu of human expert la
generation. In each generation if the aolution i1
8, EcologJcalmodtl• ,. 8. Model■ or IOdal ayateim time COlllllming and expe111lve.
Improved, It la atored u the beat aolutlon. Thia ia
OAI baa been uaed to study evolutionary upecta of ➔ Iv. Stability
repeated till convet'lence, 7, P0jdatlon genetlce moclefa
aoclal 1y1tems, such u the evolution of cooperation,
It can aui11 1 human expert in problem aolvlna and i1
8.18 Appllcatlon1 of Genetic Algorithm e. Modell of aoclal ayatema the evolution of communication, and trail-following
more likely to conaider all p0111lrilitica.
behaviour in anta.
- An effective Genetic algorithm ~preaentation and ➔ '. Effldenc7
Fl&, CU: Appllcatlona or GA Expert ayatem i1 capable of ~viewing all the
murungful fitneaa evaluation are the keys of the 9.19 Expert 1y1tem
.. 1, OpUmbatlon tranaacti0111 u compmed to human expert■•
auoceu In Genetic algorithm applicllions.
In an artificial intelligence, an expert 1y1tem i1 a
- The enlrelly of Genetic algorithms come& from their OAI have been uJed in a wide variety of optimization 9.19.3 Umltatlona of Expert System
computer 1yatem that emulatea the decilion malcing ability
limpllcity and elegance u robuat search algorithms u wb, including numerical optimization, and
of human expert. An expert system is uaed to extnct the (i) Limitatiooa of the ICIChnology
well u from their power to find out good aolutions combinatorial optimization problems such u ·traveling
information of human expert within specific domain and
quickly for difficult hipHtimeualonal problema. lalcaman problem (TSP), circuit design, job shop (ii) Difficult knowledge acquisition
makcl this knowledge available to leas experienced uses
• Oenetlo algorfthme are UNful and acbeduling and video & sound quality optimization. (ill) Expert 1y1tems me difficult to maintain
efficient through computers codes program.
whan .. 2. Automatic: Pro&rammlnl (iv) Align development coata.
OAI hu been uJed to evolve computer programs for 9.19.1 Characteristics of Expert System
l. The 1C11Cb space is large, complex or poorly 9.19.4
specific tasb, and to design other computational Appllcatlons of Expert System In Civil
undentood (i) High performance Engineering
structure&, for example, cellular automata and sorting
2. Domain knowledge i1 acarce or expert knowledge is networb.
(ii) Undentandable
difficult to encode to narrow the aearch apace (iii) Reliable
i) In environmental civil Engineaiq. one of the 11101t
. . 3. Machlae and robot learnlq important clwactfflllics ii the involvement of
3. No lllllhcmalical analysi1 ii acceuible (iv) Highly reaponsive
OAs has been U8ed for many machine- learning multidiaciplinary areu (chemiatry, biology, Ouid
4. Traditional aean:b methoda be unaucceaaful appliclliona, including cluaification and prediction, 9.19.2 Advantagn of expert system mechanics, l!Lllhemllica, atatiatics, economics. IIIIC
and protein &true~ prediction. GAs have also been
- The advantage of the OA approach ii the eaae with Aclvantagn ol expelt ayatem etc). Expert system in tbla na 111ually deal with
uJed to design neural networks, to evolve rules for
which it can handle arbitrary types of comtrainta and hazardous waste management, waler quality procection
learning classifier systems or symbolic production
objectivea; all 1uch things can be handled u weighted systems, and to design and control robots. I. Acceaslblllly and water resources,
conatituenta of the fitDcU function, making it euy to ➔ 4. Economic models ii) In stNCtWal engineering, expert 1yatema have aided
adapt the OA acheduler to the particular necessities of a H. Cona181811CY design, analysis, inapcction and maintenanee and code
OAI bu been U8ed to model processes of iMOVation,
very wide range of pouible overall objcctivca. checking. There is also • great poeential applicllion for
the development of bidding strategies, and the UI. Time OOMtralnta
emergence of economic markets. diagnoail and interpretali0111 problema. Molll of the
- OAa bu .been used for problem-aolving and for
Iv. StablMty expert 1y1tema developed in thla na involve
modeling. Genetic algorithms arc applied to many ➔ s. Immune ayatan modela
interaction wilh conventional calculua and analysi1
acientific, engineering problema, in buaineu and OAI has been U8ed to model virioua upecta of the v. Elllc:lenCY · programl, and the expert ayatem ii an aid to conceive
entcrtallllDCllt, comprising : natural immune system, including somatic mutation models and to undentand atructunl behavior. Expert
F11, C9.7 : Advmtaaes or espert llydem
during an individual's lifetime and the discovery of iyatem 11 uJed for diapoai1 of safety and Integrity of
multi-gene families during evolutionary time.
atructunl aystema, diagnOll1 of buildin~ defccta.
~
~ Construction Manapen:nt (SPPU) 9-19
1
n
~ to Artificial Intelligence Technl
~
M construction Manaaement ISPPUI 9-20 lntroductlon to Allfflc!al lnteHIQence Technlaue r