0% found this document useful (0 votes)
38 views25 pages

ML Notes Part1

The document discusses various concepts related to neural networks, including activation functions, learning rules, and the importance of linear separability in classification tasks. It highlights the role of the perceptron model and the backpropagation algorithm in training neural networks. Additionally, it touches on Bayesian inference and information gain in decision-making processes.
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)
38 views25 pages

ML Notes Part1

The document discusses various concepts related to neural networks, including activation functions, learning rules, and the importance of linear separability in classification tasks. It highlights the role of the perceptron model and the backpropagation algorithm in training neural networks. Additionally, it touches on Bayesian inference and information gain in decision-making processes.
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

pereeptron Leagng

Threshdd
Bfas

w
Net-Sm,
tutput
mmation Ai vaton
funetion funchen

.The fiet eusal nlw mode prcaphon destres

Can be
set fs claesitg dala
two
Lofm&e tfatub asses, ts a Btreight line
tannot gowits prephon eoanig 7

Mcculoch- Prtt8 made a au artitica


uaon al and ebbian Lanig rule f atihi
eishtt

otheaiton,
od
3 Net Sun (mltipl. cathi,
Sumo thoe thi
wwe w
y AActivatin funet Bascd on te activaho,
fuon, nutput
ativaton fun

Funeton, we calclate utpuk "+hehol


vale
Tu caludaled olp so actual olp ia Conpased. et
[Link] a not alleptebe
odiut te weiate
prcep
Yando
For eaeh Epochy
oiol rndd the prudelta
() Apey tre activaior functan on the weighl
a-ove t thouthael
tthe
vale, outpt
the yalu;as posite ek output Value as egve

tycataly
wna at slep t
ect)= en
Aw: ditt in weiaht that
has to be oddhd to w
ANDGod 8=04
ed to iid udatonhie blw e

2 3
idipdeut
Vahiable
ydepadeut

(a-~Xy-) (-2
2
-3 -325

: 25

3 35 (125
(1.

2O

20

Ob -m G. 25- 0:15 X5
4 5
(td f prarith

(one dipeickt vaiabe

tie ina Regenin olyen


vaiable e che
Vauabl
Binay sepasate (st Bpoabe with a
staigtht ne)

Linany sapsale
ae
a tine t0 epasda
y 3 e-Ve dat
Cay

daa
suclentuw weght ve ct
pecphon ude 4hde a
S the [Link]
oheu
eauple
but C¡u Pelto CorNese ie te
nt ineey sepeaabe
.
4secard trainig rule, caled delta suk
ae no leasly sepaabe
dolta Suue
tte
to tte tatget Conupt
avodiut Bescent, to, 4eancle the hypotein Apa e
tto thd the weghe
that b t fita the fyainig exaupes
This Sule is uotaut becius adieut descet
Provide the bas'A fa BAE PROPAGArION agoitn
not achiald
houtse neun dhethee
a diidas at Valué
fyntm Aehuaion
gNonLias
function. Aivaton Linea
may
n
(raugtht (inedy
hie
NDde
3
neuson ehhea -decrdes
ole
Fumetons
chiyat
on aase
>shit
auothe One to huma
tlgi that the
wy. mimicig brain
agorithri lkasâg dap r
Functona 2Acivat
on
)eo theu reton
ip unctin Cauot

claxotiaton prodeme becaue it


valu'
The radieut tte ohic
ite oaupropogatin proeu

achvation" "det Punton

some un ts trartd to the


natnison e Linas
olp
Limitatoss}
wse ba propagon a the
duatye of tte funchn iA Contaut ad has

bajactivan tuncion aA ued.


ige laye N
Lheas. a chivoton tun tauid
mlti laiyurd Nst
Achaton funchonf they allao back
ONOn- rea deilvatie wwd be olaka
Propngatin be caut the ontaut
e oeivatie A not
Cau Fancta

beary'Rel0
ametie Relu
(ntd)
Tanbee ELU turaz(a,) GELU
Kelueite
yfey max (o,)
SE LU
i
ut mieed
on de
apey teu dkolz Cotieuoug'ydue
(uto
eel for ee. ko dethe
yalucd Cahuous thncaphati
e
nit
runig2 esht Redued (i)
addrelld
wit Cau
be,5Nbdsue
ut ting
data te fod
on
te
fe that
l daatainang the much
on tuo dpand eyo,
dala Ogeafitig
te
esnng tree cikon frn
(Tues
de fi
Linea eprab:l
Keprat tuoo lees wi

be
kingle pauphron lau

Caunot divide te two clanes u


tey you nced a mtilyr
pectepho to e those end o probein
tuwo la ot rpatts that lay
Aepoiäla decion boundany upraalz
ina uation
epasabe
sttiria, keparab sutos tote fac that
Oanes poll
wite

siyde de cison wtace

oC)
(o,)

(o,0)

epanale
a kie ine
Hele
Y

(o) G)

(0)
thiu line Cau
Ex-0R
epouate tteu,

(o,o)
Nt a
Nw

to peaate
TD das Atth Cunmd iney
TO
we
Lineas ogorthn besed
$pevikd
Spuved eaniy
predhon
pstons tthe tasr to predict dependeud yasite
Vaduet besed on a ey fdependet vaade
(x) so, this sreAen
a linear elatin
zchná fhnds Out
blwx (|p) s y(o[p)

PPsdto means
the yaes
?t a Rthea,s greAsian pode

Numeea

2 i24

3 2

So
b
m= 1,a-7) (8-8)
y, (a-)
Caludate
(TO3)

Nafve cassfe Mgoritt


pviked Leasnig aletitbn, shicl
iclaksteetiontteme.

high-irensiaal

claui cation agoritfmd hiak help ie buclding


Machino leaining Modas rat cau mot
te fast
Preoi ctoy,
Pretabistic

cect
Sarre
Speon flteaug, Sernsutal aralya, oud olaaigrg
ron! e ocsece a Ceitain foalee

red
fruit is Secognied al afrle
Pagee means; Bay heoem
NCayes Theoremn ives ee conditinel
dbabil eveut A eu anothee eyet
B has ocutred Lialihocd

Psobabili hes fai eale

las Label
PCC)
Ci,C>
PCyes) = k , 2, P
féalae athribula
Stap cuaa a
Ctolce)
P (Redlyej' f(Coto)
produet t prior pbbil
the das label Gfa the test

Deoi vochnt Bayes Theorem


hs hypothtu p(hlo) PQD|b)* PCh) Pst.
fom ypothess PCD) Daasit
spade.
hypotteie patelH)
be the tinal Qnswe
we oeal with
heH multipk atiba `o
Eagmax Pph)-P)
atbt
PCob)PCh)
lornn inteuce
-agnnPa,a,,---an)
yev
pta,,ay--- anv) Pl;)

yeV
P (Yes) cu)
TOje

No to eah Couon.
Siupe colouk Red, yell aw P(e)=p Are)
3 PLB)
P(Rod /yeu) 54cyU)
p( yolleao)
yie Red. yes) Combinatioatin . .)
’Sinilased no.
Caldate prds abl;ea

R&nethon may be ased spcihaly


Red, eUv, DOmeite4

BP(ye x) P(su/yu)x
5
Ploorete/lyesy
P( RedNo) P(Cul No)
p(oomeAHc (No)

Selec the prtbatiy *atat hag ax yae


Noo frlance to be doutied

Wind =
5yy
f(yes) )

Hothe hee as yel S D


p(yes New Endtaue) Pyes) p(Aroy/y) p(allyes)
Ciest

P(kuogiyes )
yo taly
ye No
Suerg
ofs
Rainy
)infmaton
16oin
prdici madel sed -fa lasas
cuud eah inx
outoma
Bupewind warnig modeft
uee Labelle
daä e to rain ou walri

de cison
tree CorAuchon we needto Chak a feale a
atriber
attibcte at
bet pting ats
Hto to Chooe
each lsuel?

intdu to undutaud Irnfmatin Gain


to urditaud shat is Erhopy 9
Ertopy t is a malioe

clas, Paenade
Pinode
nodr

Erty(yu-4, o= o) =
Hhen Ertpy

Enteey (S)=

dals.

Eropy (s)
Erer
(-ous)
)

) Entmotion Goin,

obt thatnode
erty

(S
Gato ls,A)= ,Enhoy (s)
Kubeet bette

Gai
Gain(sgti Enhpy(s)-Enhop (Si)
(S)

feae vort ainum fufamatin


ottibutt
ut look
Vainy
-tuie athbt
windy

NO
NO
Frday (4)

(atibta)

attbt

Solf.t dakset futo mcbek bakscn

fease
(0) the
et the
w)-Pspeot
dthised teee
de cisin tsee, aud
.
bonc out each pokbe vale beat
ND
5
ye
lg.(%,)

3
Ouaakt

E( utlor-rainy)= -lo)-l4 (/s)

(fa outtooe)
2(outto) 'Gny E(0=s) +
Pn

Pen
I(ntoc) 3+2
9+5

4f5
6693

lawpualae
Tempaal P n

hot

(op
2f!

4
95

Nfw windyP
3

3f3
95

outtoor

Thay Supat the an


Out put s bise be tane we cshe

Tut Tyainntd
pokt

alesuadi
the daaet

nCorpdatig Valed aitihia


onvet te disoct
ontinuos yaluer to diknt

Co
Appicaionds

ee
tie
9
Taindogt.
onGpt
by are lean toOhine ma os rain we
atibut
atobut
talculat
4) enp
alaat (a)
daacet
agithm
rds

(te|ye)

do noti
ee

(i Suey with Roig (

You might also like