KWEEZIJOSEPH,MAY19/ BCS/004U

CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021

SECTI
ONA

Quest
ion1

Usi
ngar
ealwor
ldscenar
ioi
nyourcount
ry,
expl
aint
heConceptofAr
ti
fi
cialI
ntel
l
igence
andexpl
ainf
ourappl
i
cat
ionswher
ethi
sconcepti
sbei
ngused.(
6Mar
ks)

AIi
st heabil
it
yofamachinet odi
spl
ayhuman-
li
kecapabi
l
iti
essuchasr
easoni
ng,
l
ear
ning,
planni
ngandcr
eat
ivi
ty.

I
tsappl
i
cat
ionsi
ncl
udet
hef
oll
owi
ng;
-

 Smarthomes,ci t
ies and i
nfr
astr
ucture:Smartthermost
atslear
nf rom our
behavi
ortosav
eener gy,whi
l
edevelopersofsmartci
ti
eshopetoregul
atetr
aff
ic
toimprov
econnecti
vi
tyandreducetraf
fi
cjams.

 Onlineshoppingandadvert
ising:Arti
fi
ciali
ntel
li
genceiswi del
yusedtoprovi
de
personal
izedrecommendationst opeople,basedf orexampleonthei
rprevi
ous
searchesand purchasesorot heronli
nebehav ior.AIishugelyimpor
tantin
commer ce:opti
mizi
ngproducts,planni
nginvent
ory,logi
sti
csetc.

 AIin Ast
ronomy ;Art
ifi
cialIntel
l
igencecan beveryusefulto sol
vecompl
ex
univ
erseproblems.AItechnologycanbehelpfulf
orunder
standi
ngtheuni
ver
se
suchashowi tworks,
origin,
etc.

 AIinHealt
hcare;Inthelast,f
ivetot enyears,AIbecomi ngmor eadvantageous
fort
hehealt
hcareindustr
yandgoi ngt ohav
easi gni
ficantimpactonthi
si ndust
ry.
Healt
hcareIndustr
iesareapplyi
ngAIt omakeabet terandfasterdi
agnosisthan
humans.AIcanhel pdoctorswithdiagnosesandcani nform whenpat i
entsare
worseni
ngsot hatmedicalhel
pcanr eachtot hepat
ientbeforehospi
tal
izati
on.

 AIinGami ng;AIcanbeusedf orgami

ngpurpose.TheAImachi
nescanplay
st
rategi
cgamesl ikechess,wher
ethemachi
neneedstothi
nkofal
argenumber
ofpossi
bleplaces.

 AIinFinance;AIandfi
nanceindust
riesarethebestmat chesforeachother.The
fi
nance industr
yisi mpl
ement i
ng automation,chat-bot,adapti
veintell
i
gence,
al
gorit
hm tradi
ng,andmachinelear
ningint
of inanci
alprocesses.

 AIinDataSecur
it
y;t
hesecuri
tyofdat
ai scruci
alforev
erycompanyandcy
ber-
at
tacksaregr
owingver
yrapi
dlyi
nthedigit
alworl
d.AIcanbeusedtomakeyour
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cialI
ntel
li
gence
Fi
nalExam 2021
datamoresafeandsecur
e.Someexampl essuchasAEGbot ,AI
2Pl
atf
orm,ar
e
usedtodet
erminesof
twar
ebugandcy ber-
att
acksinabet
terway.

 AIinSocialMedi a;Soci
alMediasit
essuchasFacebook,Twi t
ter
,andSnap-chat
containbil
li
onsofuserpr ofi
l
es,whichneedtobestoredandmanagedi nav ery
eff
ici
entway .AIcanor gani
zeandmanagemassi veamountsofdat a.AIcan
analyzelotsofdat atoidenti
fythelat
esttr
ends,hash-
tag,andrequi
rementof
di
fferentusers.

 AIinTr avel&Transport;
AIisbecomi nghighlydemandingfortr
aveli
ndust
ri
es.AI
i
scapabl eofdoi ngv ar
ioustravelrelat
edwor kssuchasf rom makingtr
av el
ar
r angementtosuggest ingthehotels,fl
i
ghts,andbestroutestothecust
omer s.
Travelindustr
iesareusi ngAI -
power edchat-botswhichcanmakehuman- l
ike
i
nteractionwit
hcust omer sf
orbetterandfastresponse.

 AIinAutomoti
veIndustr
y;SomeAutomot i
veindustri
esareusi
ngAIt oprovide
vi
rt
ualassist
anttot heiruserforbett
erper formance.Such as Tesl
a has
i
ntr
oduced Tesl
a Bot,an int
ell
i
gentvi
rtualassistant
.Vari
ous I
ndustri
es are
cur
rent
lyworki
ngf ordevel
opi
ngself
-dr
ivencarswhi chcanmakey ourjour
ney
moresafeandsecure.

 AIinRobotics:Art
if
ici
alInt
ell
i
gencehasar emarkabl
er ol
ei nRoboti
cs.Usuall
y,
gener
alrobotsareprogrammedsucht hattheycanper form somerepeti
ti
vetask,
butwit
ht hehelpofAI,wecancr eateintell
igentrobotswhichcanper f
ormtasks
witht
heirownexper i
enceswithoutpre-programmed.Humanoi dRobotsarebest
examplesforAIi nroboti
cs,recent
lythei nt el
l
igentHumanoi drobotnamedas
Eri
caandSophi ahasbeendev el
opedwhi chcant al
kandbehav el
ikehumans.

 AIinEnt er
tai
nment;wear ecur
rent
lyusingsomeAIbasedappl i
cati
onsinour
dail
ylif
ewi t
hsomeent er
tai
nmentservi
cessuchasNet f
li
xorAmazon.Wi t
hthe
help ofML/ AIalgori
thms,these ser
vices show t
he recommendati
ons f
or
programsorshows.

 AIinAgri
cul
tur
e;Agr i
cultureisanareawhi chrequi
resvari
ousresources,labor,
money,andti
mef orbestr esul
t.Nowaday 'sagri
cult
ureisbecomingdigital
,and
AIisemergi
nginthisfi
eld.Agricul
tur
eisapplyi
ngAIasagr i
cul
tur
er obot
ics,soli
d
andcropmonitor
ing,predicti
veanalysi
s.AIinagri
cult
urecanbev eryhelpfulf
or
far
mers.

 AIinE-commerce;AIisprov
idingacompetit
iveedget
othee-commerceindust
ry,
anditisbecomingmor edemandingi nthee-commercebusi
ness.AIishelpi
ng
shopperst
odiscoverassoci
atedproduct
swithrecommendedsize,
color
,oreven
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
br
and.

 AIineducati
on:AIcanautomat
egradingsot hatt
hetutorcanhavemor eti
met o
teach.AIchat
-botcancommunicat
ewi thstudent
sasat eachingassi
stant
.AIin
thef ut
urecanbewor kasaper sonalv i
rt
ualtutorf
orst udent
s,whichwillbe
accessibl
eeasil
yatanyti
meandanypl ace.

 Digit
alpersonalassi
stants:Smar
tphonesuseAItoprovideservi
cest
hatar
eas
rel
evantandper sonal
izedaspossible.Vi
rt
ualassi
stant
sanswer i
ngquest
ions,
provi
dingr ecommendationsandhelpingorgani
zedail
yr out
ineshav
ebecome
ubiquit
ous.

 Machinetranslati
ons:Languagetranslat
ionsoft
ware,eitherbasedonwri
tt
enor
spokentext,reli
esonar ti
fi
ciali
ntel
li
gencetopr ov
ideandi mprovetr
ansl
ati
ons.
Thisal
soappl i
est ofunct
ionssuchasaut omatedsubti
tl
ing.

 Websearch:Sear
chengineslear
nfrom thev
asti
nputofdat
a,pr
ovi
dedbyt
hei
r
user
stoprovi
derel
evantsear
chresul
ts.

Quest
ion2

Usi
ngappr
opr
iat
eexampl
esexpl
aint
het
hreebr
oadcat
egor
iesofLear
ningi
nAgent
s[3
marks]

1)SupervisedLear ning
Inthi
st ypeofl ear ning,theagenti ssuper visedinev erymeansi npr i
oritself.
Whati tsimplymeansi sthatthecor rectanswerf oral mosteachexampl e
probl
em i sfedi nt heKnowl edgeBaseoft hesy stem i nitiall
yi nitsdev elopment
phase.So, whenev ertheagentconf rontsapr oblem, ittriest of i
ndt hesame
probl
em orasi mi l
arpr oblem i
nitsknowl edgebasewhosesol uti
oni talreadyhas
embeddedi nitssy stem.I ft
hepr oblem i snott her eori sal otdiffer
entf rom
thosealreadyr esi dinginitssystem, thenint hosecases, t
heagentf ailsto
functi
onorper for m anynecessar yact i
on.
2)Unsuper vi
sedLear ning
Intheunsuper v i
sedl earningagent s,theanswer stot hepr obl emsar enot
avail
ablewi t
ht heagenti nadvance.I nt hist ypeofl ear ning, theagenthast oitsel
f
fi
ndthesol utiont ot heproblem byl earningf rom i t
spastact ionsandexper iences.
Howev er,ther equi redinformati
onwhi chf ormst hef oundat ionoft heKnowl edge
Baseispr ovidedt ot heagentinitsdev elopmentphase, buti thast of i
ndt he
solut
ionsbyi tsel f.Ithast heabi
litytof indar el
ev antsol utiont othosepr oblems
alsowhi chtheagenthasf acedfort hef ir
stt imeandhasnopr iorknowl edgeor
experi
encer egar dingit.
 KWEEZIJOSEPH, MAY19/ BCS/004U
CS203Ar t
ifi
cialIntell
igence
FinalExam 2021
3)ReinforcementLear ning;
InReinforcementLear ning,t
heagenti sgi vensomer ewardoccasionall
yfor
compl eti
nganyt ask.Her e,
thegoal oftheagenti st ogetthemaxi mum ofsuch
rewards.So, whenev eranyagenttriestof indt hesolutiontoanyproblem, i
t
searchesf oranalternati
vewhi chwoul dgiv ehimt hemaxi mum rewar dpoints.
Thistypeofl ear
ningnotonl ymakest heagentsmar tbutalsohelpsittotakethe
bestpossi bl
edecisionaccor di
ngtot heut il
ityofthedev el
operortheuser.The
uti
li
tybasedagent suset hi
st y
peofl earningi ntheirsystems.

Quest
ion3

Wit
hexamplesdescri
bet
heenv
ironment
sbel
ow andt
heagent
sthatoper
atei
nsuch
env
ironment
s(3marks@)

a)Si
ngl
e-agentv
s.Mul
ti
-agent
.

An environmentconsisting ofonlyone agentissaidt o be a si

ngl
e agent
envi
ronment .Apersonleftal
oneinamaz eisanexampleofsingl
eagentsystem.
Anenv i
ronmentinvolv
ingmor ethanoneagentisamul t
i-
agentenvi
ronment.The
gameoff oot
bal
lismultiagentasiti
nvol
ves10player
sineachteam.

b)Epi
sodi
cVs.Sequent
ial
.

Inanepi sodicenvir
onment,anagent'
scurrentacti
onwi l
lnotaffectaf ut
ure
acti
on,whereasinanon- episodi
cenvi
ronment,anagent '
scurrentacti
onwi l
l
aff
ectafutureacti
onandisalsocall
edthesequent
ialenv
ironment.Examplesof
epi
sodicareexpertadvi
cesystems–anepisodeisasingl
equesti
onandanswer
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
Quest
ion4

Whati
sthedi
ff
erencebet
weenaGr
eedySear
chandA*sear
ch?[
2mar
ks

Best
-f
ir
stSear
ch(
Greedysear
ch)

Abest-f
ir
stsear chisageneralapproachofinfor
medsear ch.Here,anodei sselect
ed
forexpansionbasedonanev aluati
onfuncti
onf (
n),wheref (n)inter
pret
st hecost
esti
matev alue.Theeval
uat
ionf unct
ionexpandsthatnodef i
rst,whichhasthelowest
cost.Acomponentoff (
n)i
sh( n)whichcarri
estheaddit
ionalinfor
mat i
onrequir
edf or
thesear
chal gori
thm,i
.e.
,

h(
n)=est
imat
edcostoft
hecheapestpat
hfr
om t
hecur
rentnodent
othegoal
node.

Whi
l
e

A*Sear
chAl
gor
it
hm

A*sear chisthemostwi delyusedi nformedsearchal gor

ithm whereanodeni s
eval
uatedbycombiningvaluesofthef unct
ionsg(
n)andh( n)
.Thef unct
iong(
n)ist
he
pathcostfrom t
hestart
/i
niti
alnodet
oanodenandh( n)istheesti
matedcostoft
he
cheapestpathf
rom nodentothegoalnode.Theref
ore,wehav e

f
(n)
=g(
n)+h(
n)

Wher
ef(
n)i
stheest
imat
edcostoft
hecheapestsol
uti
ont
hroughn.

So,
inor
dert
ofi
ndt
hecheapestsol
uti
on,
tryt
ofi
ndt
hel
owestv
aluesoff
(n)
.

Quest
ion5

AshokandBeat ri
cewouldliket ogoonadat e.Theyhav etwooptions:aquickdinnerat
Wendy ’s,ordancing atPr avda.Ashok fi
rstchooses wher et o go,and knowi ng
whereAshokwentBeat r
ice also deci
de wheret o go.Ashokpr efersWendy ’
s,and
Beatr
icepref
ersPravda.A playergets3outhi s/
herpr ef
erreddat e,1outofhi s/her
unpr
ef err
eddat
e,and 0 ift heyend up atdi f
ferentplaces.Allt hese ar
e common
knowledge.Usi
nggamet heor y,f
indasubgame-perfectNashequi
librium.[
5mar ks]
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
Quest
ion6

Descri
bet
heTur
ingTestappr
oachascommonl
ysubscr
ibedt
obyar
ti
fi
cialagent
s[5
marks]

A Tur
ingtestapproachi
nvol
v esaut
omati
onofacti
vit
ieswhichcanbeachi
evedby
humanthi
nkingsuchasdeci
sionmaki
ng,
probl
em sol
vi
nglear
ning.

Turi
ngt estappr
oach;i
sat estt
hatamachineperfor
mst ofi
ndouti
fitcandif
ferenti
ate
from humansanditinvol
vesnatur
all
ear
ningprocess,machi
neandi
tsdeceptiv
esi mple
methodofdet er
miningwhetheramachi
necandemonst rat
ehumanintel
li
genceinclude
robots.

Quest
ion7

Descr
ibet
hemet
hodol
ogyofK-
meanscl
ust
eri
ng[
6mar
ks]

K-meansalgorithm i
sani t
erat
ivealgori
thm t
hattr
iestopart
it
ionthedatasetint
oKpr e-
dist
inctnon-overl
appingsubgroups,whereeachdatapointbelongstoonlyonegr oup
andassignsdat apointst
oaclustersuchthatthesum oft
hesquar eddi
stancebetween
thedatapointsandthecluster
’scentr
oidisatthemini
mum.

K-
meansmet
hodol
ogy

1) Speci
fynumberofcl
ust
ersK.

2) I
nit
ial
i
zecentroidsbyfi
rstshuff
li
ngthedatasetandt
henr
andoml
ysel
ect
ingK
datapoint
sforthecent
roidswit
houtr
eplacement.

3)Keepi
ter
ati
ngunt
ilt
her
eisnochanget
othecent
roi
ds.

Comput
ethesum oft
hesquar
eddi
stancebet
weendat
apoi
ntsandal
lcent
roi
ds.

Assi
gneachdat
apoi
ntt
othecl
osestcl
ust
er(
cent
roi
d).

Comput
ethecentr
oidsfort
hecl
ust
ersbyt
aki
ngt
heav
erageoft
heal
ldat
apoi
ntst
hat
bel
ongt
oeachclust
er.
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
Quest
ion8

Di
scuss t
he di
ff
erence bet
ween Hi
erar
chi
calcl
ust
eri
ng and K-
means cl
ust
eri
ng.
[2
marks]

InHierar
chi
calcl
ust
eri
ng,t
hetwomostsimi
larcl
uster
sarecombi
nedt
oget
herand
cont
inuetocombi
neunti
lal
lobj
ect
sar
eint
hesameclust
er.

Whi
l
e

Thek-meansal
gor
it
hm ispar
amet
eri
zedbyt
hev
aluek,
whi
chi
sthenumberofcl
ust
ers
thaty
ouwanttocr
eate.

Quest
ion9

Descri
betheprobl
emsanagenti
sli
kel
ytoencount
eri
napar
ti
alobser
vabl
esear
ch
space(2Marks)

Explor
ati
onpr
oblems:Whent hestatesandact
ionsoftheenv
ironmentareunknown,
theagentmustactt o di
scoverthem.Expl
orati
onprobl
emscanbev iewed asan
extr
emecaseofconti
ngencyprobl
ems

Quest
ion10

Noneoft
heVar
iabl
edomai
nschange

1={
B} 2={
R} 3={
B}

4={
R} 5={
B}
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021

SECTI
ONB

Quest
ion1

1.Thetabl
eaboveportray
sadmissionrecor
dsforst
udentswhodidPCM combi
nati
on.
Thegoali
stobuil
daNaï veBay
esCl assi
fi
ergi
venthedata.What
’st
hetheor
ybehi
nd
Naï
veBayesClassi
fi
ers?Whyisitcall
ed‘Naï
ve’
?[4marks]

Thetheor
ybehi ndthecl
assifi
erassumesthatthepresenceofapar
ti
cul
arf
eat
urei
na
cl
assisunrel
atedtothepresenceofanyot
herfeatur
e.

And i
tiscall
ed Naï
veisbecausetheclassi
fi
cat
iontechniqueisbased onBay
es’
Theor
em wi
thanassumpt
ionofi
ndependenceamongpr
edictors.

2.Calcul
atethepar
ameter
soft
hemodel
-par
ti
cul
arl
yP(
Admi
tt
ed)
,P(
NotAdmi
tt
ed)
,
P(
B|Chemist
ry)
,P(
C|Mat
h)?[
8mar
ks]

Fr
om t
het
abl
eint
hequest
ion;
-

Tot
alnumberadmi
ssi
ons=10

Admi
tt
ed=5

NotAdmi
tt
ed=5

P(
Admi
tt
ed)

P(
Admi
tt
ed)=P(
Admi
ssi
ons|
Admi
tt
ed)
/Tot
alnumberadmi
ssi
ons

=5/
10

=0.
5

P(
NotAdmi
tt
ed)

P(
Admi
tt
ed)=P(
NotAdmi
ssi
ons|
NotAdmi
tt
ed)
/Tot
alnumberadmi
ssi
ons

=5/
10

=0.
5

P(
B|Phy
sics)

Tot
alnumberoft
hoseadmi
tt
edwi
thi
nphy
sicsB=2

=2/Tot
alnumberadmi
ssi
ons
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
=2/
10

=0.
2

P(
C|Chemi
str
y)

Tot
alnumberoft
hoseadmi
tt
edwi
thi
nchemi
str
yB=4

=4/Tot
alnumberadmi
ssi
ons

=4/
10

=0.
4

3.Giv
enanewstudentwhoget
sCBA.Whati
sthepr
obabi
l
ityt
hatt
henewst
udentwi
l
l
beadmit
ted?[
8marks]

Gi
vent
henewst
udent
,thenewnumberofadmi
ssi
onsi
sseent
oincr
ease

Thi
swi
l
lbe,
=11

Thenewpr
obabi
l
itywi
l
lbe;
-

=P(Admissi
ons|Admi t
ted)
/Totalnumberadmi
ssi
ons)+(
P(NotAdmi
ssi
ons|Not
Admi
tt
ed)
/Totalnumberadmissi
ons)

=(
6/11)+(
6/11)

=0.
5+0.
5

=1
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
Quest
ion2

(
i)
.

Ul
ri
chi
slooki
ngf
oragr
oupoff
ri
endsf
orhi
sst
art
-up,whi
chdev
elopsandpr
ovi
des
someweb-
basedp2pdownl
oadi
ngsol
uti
onst
ocol
l
egest
udent
s(t
hisi
sbef
oret
he
l
awsui
ts)
.Ul
ri
chhasdet
ermi
nedt
hatheneeds2C#Pr
ogr
ammer
s,2Fl
ashDesi
gner
s,1
Phot
oshopGur
u,1Dat
abaseAdmi
n,and1Sy
stemsEngi
neer
.

Assumet
hati
faper
sonknowst
wol
anguages/
sof
twar
es,
heorshecant
ake
ont
wor
olesi
nthecompany
.SoAr
thur
snar
roweddownhi
ssel
ect
ionst
othe
f
oll
owi
ngpeopl
e:

Name Abi
l
iti
es

Jer
emy Phot
oshopandFl
ash

Br
uce Phot
oshopandFl
ash

James Phot
oshopandFl
ash

Jane C#andDat
abase

Mar
y C#andFl
ash

Pet
er Sy
stemsandC#

Ol
i
via Fl
ashandSy
stems

SupposeAr
thurknowsC#,andonl
yhasf
undst
ohi
ret
hreemor
epeopl
e.Modelt
his
scenar
ioasaConst
rai
ntSat
isf
act
ionPr
obl
em -(
usi
ngv
ari
abl
es,v
aluedomai
ns,and
const
rai
nts)
.[8mar
ks]

Sol
uti
on
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
C#Pr
ogr
am.
..
..2(Ar
thur
,Jane)

Fl
ashDesi
gn.
..
.2(Mar
y,Ol
i
via)

Phot
oshopGur
u..
..
..
.1(
Mar
y)

Dat
abaseAdmi
n.,
.
.1(
Jane)

Sy
stem Engi
neer
.….
1(Ol
i
via)

Ar
thurwoul
dchooset
hef
oll
owi
ng(Mar
y,JaneandOl
i
via)because;

Mar
ycandobot
hC#andf
lash

Janecandobot
hC#andDat
abase

Ol
i
viacandobot
hFl
ashandSy
stem.

Descr
ibet
heheur
ist
icsy
ouwoul
dusei
nthi
spr
obl
em t
oimpr
oveef
fi
ciency[
4mar
ks]

Backt
racki
ngsear
chal
gor
it
hmsar
eof
tenusedt
osol
vet
heConst
rai
ntSat
isf
act
ion
Pr
obl
em (
CSP)
.Theef
fi
ciencyofbackt
racki
ngsear
chdependsgr
eat
lyont
hev
ari
abl
e
or
der
ingheur
ist
ics.Themostcommonl
yusedheur
ist
icsar
ehand-
craf
tedbasedon
exper
tknowl
edge.

Tocapt
uret
hecompl
exr
elat
ionsamongt
hev
ari
abl
esandconst
rai
nts,wedesi
gna
r
epr
esent
ati
onschemebasedonGr
aphNeur
alNet
wor
kthatcanpr
ocessCSPi
nst
ances
wi
thdi
ff
erentsi
zesandconst
rai
ntar
it
ies

Descr
ibet
hemet
hodol
ogyusedi
nDeci
sionTr
ees[
3mar
ks]

•Var
iabl
esel
ect
ion.Li
kest
epwi
sev
ari
abl
esel
ect
ioni
nregr
essi
onanal
ysi
s,deci
sion
t
reemet
hodscanbeusedt
osel
ectt
hemostr
elev
anti
nputv
ari
abl
est
hatshoul
dbe
usedt
ofor
m deci
si
ont
reemodel
s

•Assessi
ngt
her
elat
ivei
mpor
tanceofv
ari
abl
es.Onceasetofr
elev
antv
ari
abl
esi
s
i
dent
if
ied,r
esear
cher
smaywantt
oknow whi
chv
ari
abl
espl
aymaj
orr
oles.Gener
all
y,
 KWEEZIJOSEPH, MAY19/ BCS/004U
CS203Ar t
ifi
cialI
ntel
li
gence
FinalExam 2021
v
ari
abl
eimpor
tancei
scomputedbasedont hereduct
ionofmodelaccur
acy(
ori
nthe
pur
it
iesofnodesi
nthet
ree)whent
hev
ari
abl
eisr
emov
ed.

•Handl
i
ngofmi
ssi
ngv
alues.Acommon-buti
ncor
rect-met
hodofhandl
i
ngmi
ssi
ng
dat
aist
oexcl
udecaseswi
thmi
ssi
ngv
alues;
thi
sisbot
hinef
fi
cientandr
unst
her
iskof
i
ntr
oduci
ngbi
asi
ntheanal
ysi
s.

Quest
ion4

I
.

Alearningagentisonewithlearningcapabi
l
iti
esbasedoni t
spastexperi
ence.
ForExampl e,SIRIi
niPhonesr espondstoafterlear
ningfrom i
tsenvi
ronment
andrelatestotheprevi
ousinformationtogi
vef eedback.

Al
ear
ningagenthasmai
nlyf
ourconcept
ual
component
s,whi
char
e:

 Learning el
ement
:Iti sresponsi
blef
ormaki
ng i
mpr
ovement
s by
l
earningfrom t
heenv
ironment

 Crit
ic:Learni
ngelementt
akesf eedbackfrom cr
it
icwhichdescri
bes
how wel ltheagentisdoing withrespecttoaf ixed per
for
mance
standard.

 Per
for
manceel
ement
:Iti
sresponsi
blef
orsel
ect
ingext
ernal
act
ion

 Probl
em Gener
ator
:Thiscomponentisresponsibl
ef orsuggest
ing
acti
onst
hatwil
ll
eadtonewandi
nformat
iveexper
iences.

I
I. Ut
il
it
y-basedagent
s.

Funct
ionUTI
LITY-
BASED-
AGENT(
per
cept
)ret
urnsanact
ion

persist
ent :stat
e,theagent’
scurrentconcepti
onoft heworldst
atepossi
ble
states,possiblestat
esthatmaymaxi mizehappinessrul
es,
asetofcondi
ti
on-
acti
onr ulesaction,t
hemostrecentacti
on,ini
ti
all
ynone

st
ate← UPDATE-
STATE(
stat
e,act
ion,
per
cept
,possi
blest
ates)

r
ule← RULE-
MATCH(
stat
e,r
ules,
possi
blest
ates)
 KWEEZIJOSEPH,MAY19/ BCS/004U
CS203Arti
fi
cial
Intel
l
igence
Fi
nalExam 2021
act
ion← r
ule.ACTI
ON

r
etur
nact
ion

whi
l
e

Model
-Based–Agent
s.

f
unct
ionMODEL-
BASED-
AGENT(
per
cept
)ret
urnsanact
ion

per
sist
ent
:pr
ecept
s,asequence,
ini
ti
all
yempt
y

t
abl
e,at
abl
eofact
ions,
indexedbyper
ceptsequences,
ini
ti
all
yful
l
yspeci
fi
ed

appendper
ceptt
otheendofpr
ecept
s

act
ion← LOOKUP(
precept
s,t
abl
e)

r
etur
nact
ion

I
II
.