wite the difeence mong arttfeial trte lligence.

,eng
nachioe leaning, deep leanhg
A s a feld o f 4subset of AI thal
Decp leomig
Asubsel of machire
cooputer Science-fou involues -bang ah-leanioq that uses
Sed on caeatg Saskems-orithms-to leorni patte neural netwosks
copsble of pesfemogn fomi dalad make
mata ofth man laqeu to
asks thal Yequíe decisfoo based oo mode) compler patkn
hunan fotelrgence-hal dala n large dalasels.
ncom passes MI,DL, Schudes algorihms Spectalaed neusal
NP, Tobotics, ete for leaning kono dola nlo brsed algothin
Aeans atfems kens fo laqe
Minfes human 4mount of data
frorm data. singneusa|nela

|tlgesilhms equfre nlgoftfns ave dhected Alzovifans are

ovesstyht Prob
by dala analyhcs to aqely self dhected
nan -functioo e1mie specihe vaial on cbla anakaes
der to once they're put
properlyoe hri fo dala ses.
fnto production

Hyplcatons hckuce Appltakonsinclu de ypleatashelude

email-ltemg. Fnaqe neomgrita.
chatbets, se|f dyidhg
becorotehdation engines vofce assislanls.

oFten lowfntevpelabt-ighes frtepre lattlisy oco intepelRy

oDdelssodels capler
De;
-lity due to conmplerig-linSimpler
 Gjnevsly mschie eanlng system car be

&)
leanhg lgoii
) esfeeenmerlSupersed
mahine
learning
fofo erneNt
tenng

Algovfthm

machne leaniog Alqonffom,ohich

’I4 ss one of-lhe dotaSothat we-taan he
provide A siople labelled
and on he hasts of tha we can
machine easily
ctact output vesutts.
pedict
dobeled date bet wwe CanCaeate dats
-haining machine model elfcrent
to a Set tor
dmedel
1fochine learnhg Can createundesland
qeneaie
a

belled'dala to -the
by ustng a
ertive dolaset and Analyze exch< eey
-the dalase! -fhen ilofs tamed proteSsea
 poopeny coupleted vot car poulde -the
-ths &
’ Che nothing but sanple dala [Link] data
test dala nothing whether we dve
check or verily
Set 4 ue cn
expected yesutts ox not
Ihe
supervsed lesrig gortfin
’ be mao goal of data ioto output dala
frput
Is -to nap-the inves
have supesvisio eortol
classattn of Supervtsed levomg alort
’ he

) classfftalen -hqotin

’ fI is extemely the entre datasel which oe

suppyiog -to he model ( patasi ze shoud be

Test dala:
’ The Sample data rom he dota sethu
Caleulate the perfomsh ce of the model(ala
should be smaller Slze).s
Size
 Tn thfs aoe don'! have any apeniston
ale can sioply-brañ -lie machine learntng olgthm
by ustng -the vanlous dalasels -to împrove -the
pesformance but 1 s acemplex prtcess
-to -ffodoul
upevsed
the
leonhq algortm-y
of -the dala sets
e have 2 types of unsupetrd lesninq
hm alqoHf)chustevnq 4Hqorilhan
) 4secta tion leantog algrithm
Based on thase ayovthm
leditfeent com plex probie ms t s
Reirtorce ment luaminq -tlgoithm:
ohe of the

4thot is based on feed back systen ohere we

have a learning ant;no
Jhe leaming aqent can qet a atud ohen
then s posi-fve ac-tton ,nd he car qel

penalty
’ he lesny gent attemateslly fmproves he
pefosmnce -fox huge snenl 4 data elfcthely
Be mafo qoal of oefofocement alg. to to
Ids s possible. based oo
qdie as many e peotormace
ressds Atomafetly lhe
-hese
 slastks
logistic
Machine TD
mathenalt
Yesults sxpeoted the qet can wehat so
dek. ariaus oideaigned th must cetaioly
be
lesninrg suctesful desiagn -toTrader
prope kesinng syotem
y ehey -le-tabihg poene
cheesng the-anqet -tunc hion
3) chocsig the rpresenlatan of -torqet Aucho
) chocsy he approzineton of taqet tneion
) hal deslan
cheesing- the-taifng erpevience
|choosingthe-argel-tunetron

chesfng the epsesenlaio

lagetfurttfor

choncir - he paoxínatin f
aqettunetion
 O chosing the-batotng exposience
fshefhnst step-to certfbute tbree
Thfs
athibuk - y fee Hback
a) Deqree
3) 0tsti baten eramples
y feedhack!
nhs e need to make sure -hal chethes
-hr -bainfng expevence pmovtder etfher drect teedbo
oy fndirect feedback:

By considerinq the task oe need to Pfod the

CP)
pesfomance fack fIt is qod. ther obuí obvi ously
-then oe cdn have a qod expetence (E)
(Task J
-OoddO
Letus consîder checkers qame
d
7ame 4 leamtg dlitnq
eystem
checkers game properties!

Bloct
pfece
9-An elenNent alg moves în fonusnct dve clton, i!
Contafns coly dagona"y
* 94 wil move foraYd (onley ene step) ata Hme.
x T tofll not come back
TI wtn fump ove
Oves An Pponenl

Dhrect:
sheo ue plaq a oheckess game to drect mode
Autom>teally the system aotl qve suggeobons
end dire c-hons to
play
Ipdirect
no one can help fn fodhect made. Nlosagpert
tom system sde.
Aeavning Drtdng in dnecl mode
nlhen oe are leanfng'tafulog
theeSho uld be 9 tatness

be sny support -fom

These sho euld bet
tafoes

taiofog expevience ehoes -the

ferehe
equence of dhing
 fs as -folloues

equene

On thef oon
fou take -the
tafner's help
partally
we toe the
of bafoes
help of

sfhuttbns shieos2
y Jumber d
2) nlum ber of possfbilhhes
) tembes ef etomples OVer

Peneton
o!
’ Tasqet func Hon focludes -the suces rate of
-Ihe qiven algo ffim hnsa s
’ Tht deseibes aboul legal moves checkess
fome eampk
movts nothing but nor4 pos?ble
 |Legal move,/
one moye

Succ e ss

Checkes's
Tavel only in -foruand decto
One mNOVE per dislance

only to dfago)dreeto
Jump Der oPponeol
funeHon-v(b bar
br BoAi slate -b

’b is the boArd slate, thet ison theo v(b)- po

--1s
’bsthe boasd state, thal fs losl then v(b)
slate,-hal is dao then v(b)-o
’b s he boad

6'- bestfoal slat.

 furc Hon:
de
(b)means for
boayd slate fuoctto '
caleulate Ihe
lasng statr, oe bond fanetien featu
Bo ton
a of folocong
Ihear Com
c(b).
feahves:
bl¡ck pieces on bovd
2|= no-
no-e4 red pfeces on boand
, =
black kfogs
on boárd

5- 0- back pfeces thestened by ved

(blacks ohich Can be (thnelen) bester by re d)
s - no-cf red pieces hea lened by kack.

(aepresentahon c(b).
g- c0g - dumevical coeficients (or)
teatuse.
 OAdiive conslnl

the Iavqel func hoo

ion:
o leon atarqet func tHon fJ
( voe need a et
sel
of -thofoling
ag
ezarples
fx descaibe a parteulay boavd slate
A-taiofoq
value) Vhaf (b)
|ondered pa'r - (b Vato (b))
Trafoing ciampe vepsenla tor
fu Black con heone
E
eg
je )-0 , ohfch means no e d
black
preces

We need to do -tuo0 steps fo his phase.

 alues
Consder suctes50r
woe
stef
mert sep of ponent)
on -he
Qafo (b)¢-v(sitesso (5))

next bond slste.

Tepents
(kstlimating that -ths meve l help/destby oppiney
v'-eprsenls pozlrafon

hese ae Some of

lineay functons, here we are Lms

keast Meon Square used to fodd dnd ofoienize

he en0r

nr(vtailb) -v(6)
ff fnorso. bo ocd -to chnge eights
il enoztve, eoch wegbl s incsease dh propoiton
S/ enoc-e each weight js decveased in proprsit
 ML involwg leercbing
hypothes, to
one

data and

hypehes: Enplonatn a prblem t

snefnaten.
TIsuey ML:
What aenthy houd be wed

S whok mehods shbuld be cco Tedu,