Scribd
Upload
8 views

Types of Neural Networks

Uploaded by

ankushpan694
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
8 views

Types of Neural Networks

Uploaded by

ankushpan694
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Ty pes of NewaL NetugoKS

IANN -Atifictal Newral Nenuook


muudple pekcp tonu /muwom) at each leuel.
aluo KrDun as feod to wand munal twoh
becauue thpu a papcessed omuy tn
fomwand dbuctlon.
simplest Vasdant of muwnal mutuunrk
4 may 0r maymot hae hidcun modl 1ayus

O
C
O
O ol playee

1/P1ayer

hiddun layenl hidden layn 2


Adwantages
t h u Store 1hho on enti mutuuoaK
hauue tha abhiuity to wodK wulth incompleti Kndoladge
offex fault tolexanece , hauu dlltibutd mamoy
DiSad Vantages
huge hacuwasu dependun cy
L Somtimu hau unexpectal behaiow
m o S pecific u l for datuumindn g t h stuudu
Of ANN
2) Convolutional Newal Netuuorh -CNN
Ly most Populaa model used today.
e of multtlayer pecaptOn
vouladlon
Condains onu or more convoluti on al 1ans
0r
that can eithu be etiay Comnicud
POoled
fc-3 fc4
4mom 1insa pocessing
ReLU Neuna
Conv-2 activanon Nugak
Conv Max 66)
C5X5) Kindl Poolfng Nax with
vadd padding (2x2) vaud Potfng
t22) auopout

Flotbr
02
input michanels michannels m2channela m2
(28x28x1) t4x24xm1) ti2 x12Xmt) txexni)Channul
CAX4xna
Output
Advantoages 3
wits
Ofen veuy high accLnacy fn img recogmitiom.
weight shaing
capablu 0f automaicauy dtucting imp feates.

DsadVartages3
L CNNS domot encodi thi positton and
Of the Ohject ooieniatiom
lax tha abittty to be Spaitay invaiant to input
data
a lot of kaining 1S uquuud fom it to work
esoiciunty
3 RNN RecLwwwnt Newial Netuoa

4 Lot mmau complex


Saun duzput of paote ssing nocu and feed the
u t back fnto thec modol
L do mot pass injo în ome aiucti oh.
each wdu act as a mumoy cell Comtinuing
and 1mplem of opexaiions
the comput auon
Sustem seu, leauns and Continuus wening
towaud th CoTTect prediction ding
bacKpropogaion.

AdYantages
LRimembexs each and euy into thuough t{ma/LSTM
L4 wed in Convouuti onal layes to extend ejhectiue
ptxel mighboohuod

Disadvantages
hauu qladiunt vanls hing and exploding problems.
training RNN 1s a ditpicuut task.
L Canno+ prote sS Vey long sequn ces f wing tanh
Oreuu as an acvatlon uncuon.
ANN CNN RNN
Tabulox Text Data Smg Data Seqwnte Data
NO paam etu Hes yes
Shauing
Opeot On Hxed don't wLosk 0n
Op exate on fixed
length 1nput Hxed mput
tength mput
NO sCwwnt Connecn NO ucuwunt Connue Th a posstble
thy aun poss1ble NO Spaial sulationships
NO SPaial gel0ion ships
lecut po wwpl most poulluhul RNN ncludw less feattw
compaibility.
Hauing fawt toleuante, Htgh accwaoj in Romambeus each and
ability to wook ulth img cognitiom esey ntom auon and
incomp lett Knostedge. problems and Off es time Sesuis
eight shatng. puot cHon.

You might also like