0% found this document useful (0 votes)
13 views44 pages

MLT Unit 1 &2

The document discusses various learning problems and methodologies, including reinforcement learning, classification, and Bayesian networks. It highlights the importance of designing effective learning systems and the use of support vector machines for classification tasks. Additionally, it covers different types of SVMs and kernel methods for handling non-linear data classification.

Uploaded by

devendra vikram
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
13 views44 pages

MLT Unit 1 &2

The document discusses various learning problems and methodologies, including reinforcement learning, classification, and Bayesian networks. It highlights the importance of designing effective learning systems and the use of support vector machines for classification tasks. Additionally, it covers different types of SVMs and kernel methods for handling non-linear data classification.

Uploaded by

devendra vikram
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 44

wl-Posrd leaining problems -

An aguut Arus a pooblem ol lask T, ferfrmamco Pamd


alh bemu e jena. 6.
%Pi measiad at T i t an mpnue E

(leuag speience ) --
LPRaymg eleclars roblem
Hamd. item Aaozuikon broblem
3
divin leani problem.
.Plauimg acKarp p'roblem-
Task T playing checkes
A Pesr manca meaßuus%e gamus wom
against
3 appomens.
raimin txpaliem te
playing þrach'ca gams4.
-

againah thsalt
2. Hamd wait ing
recrgnihomp blem-
KT: Recrznihion and elasiftyiny tanduwriHan
wods urik in
images
. Pex formamta Maaz
[: fercamt qf wrds Cotha elly
3
ainina, ex þenienceE 9Aechied
daa bake. ot
-

A
words unH tonduanten
giuem claMifiahon
T H leanutn, up ftel maintaig
a individulas lasntu parfurmama
kubils enlamad takiny nto
Consiolarahbm the difteut chahnels
aarCaphom

leanaiy L : avdihua. laanni by listening.od ShaKig)


Leannin TyH2 visual
learminp ("thrwgh ta asby wateig)
umui Tye3abic learii ( "by trvclins and tal
anuin 1yk:aaauig turgh thu infellec.
T kid o classiftcoahom Call4 fora cikco cnalysis leanuiu
ype L t 3 difte n t kid t T*caphu clanuf

(Snyadyveda) oram in frmah'om.

T a pudmdna leonu les viual, Adi ta,


Raad./uAfte, and Kimaesthahetile. mosd ys may
genenal tolea, abrnl hais ue lean akt
ho Acmu

cten F Comes as a lunpnke temm u disCouay


utat wdiier cm þredominant leai skyle lk.

Vival earni.
Auditoy learning
Reod/ wite. leanui
inassthe hc leaui
Ver bal ami

c a l lean,
phsical arni
Designi a lemi stm
Togut a wccoselul Leenui stem it sknild be deaizned
T a ropu doin seural stehs shnuld be oll euad

4Ate a
LChrosimg. t taaiuiug exbaliemee
2.
3.
chorsing e taagat mchon
Cheoin a reaþesendlim-fr Aanget pnchom.
4
chrsina leanning. algoi thm for approximahg
a

the Tang Jonchon


mau dearn is obi
stebl Chorsing ATraining Experema -

an chorsin a aiui uq er
poniemco, 3 at buti anLaKea
. T. ot teadack
3.
Oishibut iom dexample
ultten ta 7aii. erbememcr rouid es direct
or indirect
ferl batk jngondina, ts clhrius mada
b pevfomamta Aysem
Direct eedback
Sndirect feed back
3) Destaa o which learming untl Conhel tto saumea.o
Taii
C leapinimg- dAiving
l i t ainer3
Wit Tainar Complatelym
parha ulp
Heuo-unel iF repseteta J a disi butim o examble
Or wticL the peafrlmamta t tinal 4ystem ls
meaned
. prsi ble. Combinahom^
mo vahcms more Rxamplu.
loami or diskuhad onga f axamble
C Learmins driving

Atep Cheoin te Targa mchm-

a taerormama atem.
c e ckars 9amR
W i l e moving, dia gonall TaveR only, frw nd direchon
t a l prstble mous onome mou perchanea
Calledl legel, mous
my. in diagenaldireahon

One mo
m oe apponum
Taaeh moua

seb3 Chari a Teresendation tr Targat fnhon


For onboqrd sate, caleulat meh on 'c' as
iaar Gmbinah'm d iluin board featunas
featua
Ne blacK plie cas m bond
M o Mo ed pieces om boand.
K3 Mo black King m board
0
y No Yed Ki on boaac).
No oblack precas threatennd b Ted
(blacks wlrch eam be beaten by red)
Mo red pie cos thredemd
by blacl..
epcesemahom o ct)
(wo-wt) numnial clftaents r t g t d eac festu
dtermine the improrkama. uegktage
Peakuu p.
wo addi h'u Gmstant
step: cheriing A earming. ARgoi thm fr Apprimaing
l a torget mch o
To learn a tageh trch m (+) a u d a Set
ai Axampl
deBcribe. a panh Cular baard State amd talwiug ralu
Vhainb)
grdrea b, train (L)
Train exampl Tepresentatiom)
Bayesiam metwork - A Bayusiam metuork isq Gmpac

text ble amd inter preable. Aapnesendaliom


melul
sm prbalilik distaikuhiom. Jt is also a
am
a
im Kusuladge lisco ve4,
as olirectead cy cie taphs
Jar
hakueen vaugbles.
allyt Aebresentimg Caunal A lahioms

a u euoutt
Bayenam HutwrKs ale ioleal To aking
Ahat occweal ame pTedich'ng Ua likali teod t h a
pestble k u s n Causes w a s t
aA ome oearal
a Bayeham neturK
Con i buing fackrr. Frexample,
behuaen
Couled r e þresent e bro babi ishc Allahmsli ps
dieas coma Synpems

Lale

0 2 n lime

ver
Am
m sle
bus ove
Yes
alaTm | yes Ao yes
O.9 OL
yes o'L o 9
yes No 02
No yso3 7
No No| 0:9
o9 OL

"Proviels glablical Anprereutatinm


a d tk probalalik
Aelatonruip ammum* a set
E Medueal biagnois-

D= HeaHy

(Cxercise)
HD-7 H
Heou Hoar D-HaalH o2
D UmhanHR, o.8S
DiAa
D-0H

D=H
EN chost
DuH resun Pain CP=7
HD Y

BP Hi H:Y o 8
HD-7 0.817 HB oy
HL H
HD N 02 THD-H
Hb-y
hne a Two Key element m BBN. MD N
HbeH
DAG DirecF Acyelie aph
Arobalitil able a Nod
roperty- Cmdihomal Indabmdent
Advamlayau Bayesiam. Beliep Nutuork

O Pmi Knuledge Lass m


CoMal snechdeds modal aw
a) gF Can amdlle. youn miiny dala
Smming a mtegtainy. ufteutvatas.

n
Cansvchng.
gemercul
this DAG
rs vy Gmpley.
PCX1 Xa --Xn) T P(X1anowha (x
=
r)))
Thll gentdishikuhen the yuaph"shuchntedl apfnogebnah
Examplep -

Manginal mola.bandemte
PCA,8,c) P(h) PCB) PO
(A: D)
(B:s
Candihionall imdapendent ettecks
Pa 8,) Pl8[A) P(c |A) PCA)
B amd c an
Cmdihonally mdspeuslout
Sivem A
(A: Thaf (B tat aku
gndapendet Cunes
PCA,B,) -
PCcIA,8) PCA)
(C: t PC 9)
Explaig aun
MKov depeusum
PCM fle18) P(81) PA)
Subortvechr Machine
Ar
oVm s alinear molel fr clasificahom amd
Fka goal f tk SvM Aeguinm broblem%.
algaitm is Create tR best line ordecisim
bomdany tat Cam seqregali m dim-emsima spaca imto clae
so lhat u L Cam eastl put tha n e data point sn t
Cetec Galogety im tR Jaluu . Tha besh deciaiom bovndaly

oledaperblame.
sihua tykerplame
a

Cupork
Veet
AA
Maximum Mangiy
A Hyper plal

Nugatius hgperphnu
A
Now data

ModeR Pcadich n cutkut


Tarainina
SvM algo Cau ba
clni ficahen
ußeol
fext
f face algo dufechn, imaye
Caa gaizahm.
Types of SVM
O Lnear SVM
Mn- Linear SVH.
SSVM uppert vector Machine )
SvM a super vipeol ma chine. leatuing, algolithm
Ueol fo beth clasiicahem amad giuiom. 7haglns
aa VM Gm Jme4 orkuilina, beth A2qeiot ame
clansiaeahm broblem (mealdlA). 'svM algo m
pexormually uutl belk lineaT ly sepanable amdl na)
inean y belarnble data Auta
Tyhe o upport vect Machine he
-
au
Tuo Tyhes
Linear SVM Or
Simple SvM:- Used fer Lnear sebarablee
wthhdal 6 a lay sut Gn b
clawidied m so clame wit a
Simgle shaing kt lie
h Lat data is Gnsidaredu to
data, aud tha clas ie 1s ufenad
bejlineanly, Le parsqble
do a tte linear
SVM claifrer. t 14
Typically used r limear resremiom
amd elamIficahom broblem.

2. lom Unear SVM or Ker nal sV M2 Mon Limear SVM Used


dr Hon linesr Aebavab
da. data sat that am not be
line.
elasifred by using 9Shoig
Lime Tke
classifien used n this Case (s ALterred
a s a mimlimea SVM clasifre,
Jt Ras mute
lexi bitily for Non |inear dals becauRe mote fealunas
Ca be added to di yrplasue mateaol of 4 Two
oimeusimal shac.
SVH algo s bates tha
on
n cept o de cislem planes, wtn
yperploanas n useal to
cla a pet ¢
gin ovgi c

AA AA
A 44

A
A

Deu'sion Bounday M
lmber deciaion bnmdaly
MV
Ho ue choope tha. besh boumdang.
T beslecigion baamdaly s te one fhal has a maximm

dislomee. fm tha nuarest þrinta of fhuse. wo c l a .


NaaTest print foom tte ophmal decisin bamdany, Uat maximiz
tle dislance a Called uppertve ctur.

ols ci'sim boumdony

Subbrrt vech
AA

A
Magim

SUM maximm mangln clamifie rtha maxiaum


mMgim perplana re
Kermel SVM

HperHel
A hyperplain
A
A A
A

magin

M ma Ke um t the Kerna metod to vse data as i n t

and Trsem intb la ugpiresl tyj peceting data


Kerul is sed becaure. of tha At d mathematial funches
Ued d upport vechr Machi ne, that Jruus the windns

to manipulet ta data Tt Kerul fenchon wie


puall Gmuert te Taainia, sh d data So ttata nn
inan de cision smface Can be tansbmed Ao a
nn eqvahion in a
igu mmben of dimensin spacas.
Kern dnehi ms abbliesl
on
eunj dala mskamce for
Ahe pupote nmappinq. tte rlieymal mmlinear
obbervahms iwo a Righn -dimensiomal shata
T n obpe rvahams be ame
sebara ble. ntte tgh
dimenMonal Spet
7ypeao kernal SvM
.Lnear kernl SvM- d6Thne anu Two KeTne namad XI
he Linea. keTnal m be and x
dot produc of tha too defimesl by th
vecrov

Plymial KennelSvM
k(Xi X) (x. xatl)d
=

Here X Vectngg and d eresest t e deres, of


3.
Gaussim Kern polynummia

T 9w Gowssiom Hernu is am
yample ef aradua
basismchom Key nel. It Cam be reprelemted witta
Rgpahon
k
Cxii) ex- vllxr-j l12)
R giv S(gma tas avi al ole u t t
tRa Gauslan
Kemml. t shrulol, be.
berfprmunca of
accelduug th problem, neithar
Cauull, tuu4
met dnstfmatel. oereshimated and
9:
Exponenlial Kernal 7
Exbonewhial Keruls an
clope. related. to
IKerul Then ane also radial basis Gausiam
Tha dfenuta kerneltunchb.
hehute thue lwp yhe4 o
s that tt Sqpan ef tha Morm 8 remeo kerl
n
Eyponentual kemul
ThA. Junchon o ¬ypaneaf ehon i4
am

K(y) rp(-Il x-a)

You might also like