Convoluona eute nleto6sk. (eNN)?

Desceibe
2.
basia Components.

Convoltion al Aleta Nehoosk. (Cnn:

CNN ts
is a neluoosks
class of deep neda ypial
Used for and classtfication.
automadicay detets Tmportant teatues in
ee Rrages stng Rrtexs.
Basie Components
Convoltional
Appltes a trom tke
filtes Ckesnel)to Entoct leates
input
poo ling
|b- lays
|Reduces the Spotial
Spotia dimensions of 4he featue.
maps, often ustng pookng
C. Aetivation Runcion;
ntroduces non- Iineortty tato e
model.
d- alhy Connected layee
Connects all -fehe Rorns he
neuYons revious
to
las
of-ten
the
used for
neurons 4he nent
firal Classifícotion.
 2.

3- Explain the. Cort ep t of a Prter

Content of CNNS:

Conteat of Convoludiona Neual Netaosk a

flter is Somall matrt

of coetq hts tat
to input image
to deteet spectte patteans, Such as edges,
o
teattes othe peatues
Each deteats Specrete Peatres, and
to undero-tand
tRe leartned Pittess become
the stsuctuse nput doto.
|-ing

Recutet Neusal Netuooks CRNNS) om

feedfouot d
neusal networks?
Recustent aleutal Netoosk CeNns):
Recuskent Aleudal Netoosk Corneetions eat fosm Cycls.

allouing hern to maintain Mneno Pevious

mating tlem Suitable fov seguential data

inpds,
VIdeo.
Rucl tfme. Setes, teut
to -the net,
Pass
Rofotrnoon Ron .
moded fempost dependenctes
ohteh helps
feedfoatd Netuorts CPNS)
PANS do not have
have ycles
 +hat loop back
Tnfoimalion fouoS one direction -rom inpt to
out
ofout feedlact, bettes Sutled Po non
mating hem
Squential tasks.
5.
hat ase anstosmess, and aie thy signiftarl
tn deep ladntng?

Tians toiness are deep leainng mod els designed Pos

tee
Péocesstng dact
Sequential cdota lke tet, bt w4aout
Iirottetions of RNNS
Unltke RNNS 2 an atention mechanis m

-Bat to ocus drpferent parts of

allouos hem , teus
tee Seguenee Simul taneously hand liag
org- ang e depondence mose eff icientty
Siqntficance
eepetally models Ifke BERT and GPT,

have. sevolutoni zed notl language processrng ue

by enabling models to Pooces s massive amownts

Parallel. leading to state- of-l-art

Perfotnance. 3n ast s machina tanslaton
tent qenesation
, and 2uestion ansoeing.
 3.

aschrtectuse Ieke Letet

6- Dis Cuss Populas CAN lex et
VGGNet, Google tet, and ResNe.
lenet
One of he frst architecture designed foo
|digit fecogntion
Convoluttonad
t tsoduced bastc CAIN Cormponents

and fully
|tke layes, poolirg layes,
Connerted layers.
Alex nlet+
A deept and larges netoosk tlan Ceet
winntng Compettion.
2012 Îmage Net
vGGN
knaon fox ts Stmp ketky VGGNR uses Very Smnall Bxe)
iletess but goes Vey deep Co os lays)
GoogleleNe Cunceptten Nedoox)
,
Pntioduced the Shcepton module alouoirg te
Con Volutions
netk to choose. drlfaent tpes of
CIxl,3x3, 5) in prsallel af Coch layer, making
mose. elPicient
Resnet
of estdual Connections, Ohi
Dteoduced Corlept
Vantshtng qradtent protlem in vey
|hekp prevont tae
deepnetoosts alooing bettex acutacy
 Oe faded Recutent untts CGRus)? 4low do

they d tfler feom Ls TM32.

GRsC Gated
Reaent Unit)
A RNN tRat Stm pltfies he nt
moíe Cornplex
LsTM aschit ectento by Combining tee forget and
nput qles Shgle update
GRus are to -tiasn
fastes tiasn
and Cornputattorally less.
Enpensive an LsTMs bud achiee Compaaba pesfesmente.

Dieen Ces LsTMSt

CsTMS have -tfAsee- gots ohtle GeUs have only
-too
memory as
LsTMs malntatn Sepatate Cell, ohere
Combiner memoy
and htdden States.
- laye pexcepfon
8 Explain
-he architectwe of matt
a single
CMup) and tts advartoges oVeY -layey
Paleptton
MLP aTchitetuse ay
An MUp Consists of multiple aas. ot neins
htdden fayers an

Connected to ee nent, and

is
function to
each neuIon ses an actwalion
ntocece non- ttneortty
 |Asvanages Sirgle - laer
perteptorn
Non-ltneoy deciston boundariesi
MeP Can model Complex, non-lineoy elationships
due to its multiple layers and non- In axt
actvation functions.
Hghcs actudaay
(eantng hierarchical ptens, achiee Meps

bettr protmance more Complea datasets

Compored stnge poceptons.
to -byr
4hat graatent descent algort@a
tn tointog
does tt help
neal netoes?
Gtactent bescasts
to
an op timization algortan used
mínim?ze ee toss Ruretion by tetatvdy updig
model's porsometets.
Role.

neuooxks, goadtent descent helps

afjust tee weights and biases by Galenating tee

and propagoing ongs.

it kackusay d

netork.
 Yole. activaion functo.
16. what is netuoo8k
raintng of netal
Actntion fneion
acfvoon furetion tntoducos non
non-lineodiby
The
he neual netoxk.
a
netal netook woul d behavs
Lilce alinesr model, egardless
, to Captoe
|lays 7
nd Unable
Corn plen
pttans.
Role.

Acivaton funetins, Suel as Rew, Sigmoid

Suel
Tanh, ensuse eat ee netoor can lesn
-
hon lheas tedotionships betoeen inpts
s Capaule of
Dutputs maeing 4ee netok
mbdelng moso Cormple tarss (Rke
fnoge
, ProcesShg tte.
decogniton longuoge