Basic Models of ANN

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

45.

BASIC MODELS OF ANN

many times af neural


There ane
networ ks aval He. hch ae chsfted based on
stuctue, Date fos, Nurons uscd
layers ad their epth actoton fltes ete

The models of ANAI ae specifad by


the ttrree basic etties namely

1. Netuork Topology
Adustrert of weghts or Leavng
2. Actvoton functors

bNetor Topolo
An retwook topdg9 1s tte arrargment
a network alog oth ts nodee and conechoa
bres According to the topolgs, ANN con be
follorg kinds.
casied

Clmor
a)
hanng
only
O pertas
tayes.
the feed
we procea
unib/des a)
Singk feedback b) feed
Thee non-reurrestTts
say
ces conecton
ther. The has founnd
input toCan Aaye
ay Jodan()RecuYvert
networks
() ()
coeptThelayer a)fond
one tosony are flly ()
dayev
inputthe feed be
Nl
lctworks,
MuAlaye Sirgle
Scedback vecument Netuork,
werghted dvded conneced
rward etuork. laye
one lyers
of ito
layer. fad werghtupon
loopdsfet netork
hoving netuok feed fed
etuork: wth
direct,tom
&types. and fonwfowand
To
fuly sorwand means the Page No
all
other and
te nodes the
Classic nehoorkhetwork
words ANN sgoa ndes
of
()
network
lnput
and hanng
Tnpts Multhlauyer
Conneced

moYc
hasconceptThe
feed
the to
One
Htdder tthan
the
output toruwand
One
otput
betueenthe
laes
layerweighted
moTeelaye. As Output
netuork
of
dayer.
artpts, calLed feed
No....
Page12

forwand
hidden
thelY
hy ANN
s
Netwoh9:
Feedback B)
which
deAdrv dirctons
dyramic
eaches t
a)
reaurertl yb) ansme

node archrt
Connected type ot Re A
Useher loscd cuYrert ystem, usingfeedbock
eetne Tt sihe
9
uorks lops. They 2
state lopns.
iolerna to Tecurerl,
netorks networks. ohich
types
because Tbe networ
all the of
roerosny)
stote (
changes bs
netwovk: fowcan
ther feedback
networhs
fo wit% eaullibum. k
both simp lloing makes
all has
and nodes lest cortousyunt)
input nodeseach No.13Page
nural anetto
he
itfeedback
oces Sto
ea
I+ a
ad n¡n-
Classic netwrh may
otpt ineo patlhs,
be
2.
Connechons the
netoork.
LeamA
thrce Adust
mpts )
Seedbackortptthetl
method Tordan
categones mens

betweenthe
Leaming
of
of networ
Wergkts hdeeo
as
namey modfyirg shousn
k
Reingoneent
Unsupensed
eamigeaming
3) 2) atfiual or go
Can
neurons closed
loop
aperved ) learsng n to
the the
he
clasted
nto be
wegtsneural tolowirg input
specfed a PuzeNo
kang netostkin
oba
netwk, agan
of
lussic
deyan.
is
s
aSupesed
Leaning:
vector
iopt is
custers. mpt
There1S the nput ven
network
neural ane
he
On
desred
output
vector.
classto when natehed
vectors ANN,Io Evror
(D-ySgnal Network
Neura)
adjusted output
vectorutput
vctor.
no basts

f wnt
feedback presented AINI,
whch he undeY
gves
inpt mlar the uttlthe eorthisof
|Generator EoY
desired urder Srgnal
ts to
on pattern is type
unaupeYSed
Leamg,
response
indicats
nptutput toe
sypervised
the
actual
utput 1s Companed
output.
belongs.
ronmentpatem netork,
werghts
sigral,the Y(Actual
Outpu >
pplied,thenCombined
hetofom D No..15
Page
wtth Leanig (Dered
awtput)
ohch
the
Classic
he al the
be tthe des ved output and i4 4
to what shautel
ncorect. Here in this type of
or
netuoyk itse l4 mut discover the
Learnig, the
nput dota, ar
he
patems and featurea from
sor the npt dato te output.
the elathorn

Neural
Netuor k > y fActual Output)

Renforeent Learg
This type of learnng s sed to efore
strengt the networ k orey Soe cnte nfomaton
sinnilar to Supernsed learg
This learng proces
bowerer e mght bave vey less ifomato.
the tranig of ne twork under
During Detwork celes s o e
renfoement learnrg, the
Seedback from the ronment. The feaback
evauate not tnstuctve, hichmeans
obtaincd he re ls
SupeYsIoO.
ttere is

eceng the feedback, he netork


After
perfoms adjustmerts ot tthe weights to gt
b ter cnthe
cntic Infooton.
Paze No 1

X Neural
Nletwork y(Actuol Dtut)

Error
Sgral Error
Signal R
Generotr (Rerngorcerent s )

3. Actvaton fnctons:
An actvahon uncto 5 matheatal
equatton that detemines tte autpt each
element perceptron neuron) n te neural
netusrk.
Tt ta kes input from each neuron
t ito as outpt, usually
and transfos
between and 2er between -| and l.
be defrcd as he extra ferce
may
over the ' n t to obtain
OY <ffort oppied
an exact outprt.
the actvahon funtons over the
To ANN,
inpt to get e exact autput

Some of the achvaton functons ane:

) Linear Actvaton funchon


also called tthe dertty Sunco
as it peforns no 1nput cdrtg. Tt can be
defned as F(*) X.

Classic
Page No......8 ****.,

() Sigmoid Actvathon functon


inay Sgmad unchon: This actvaton
tuncton perfoms nput edhrg btwoen O
and |. posrtve in natne. Tt can

be defned

fl«) sgmk) =(l + exp(-x) (te) =sgnl)

Brpalan sigmYicat fiuncton: This achvathon


Juncton pesforms nput ed'tnq behueen -)
and l. Tt Can be past1ve negahve
notue. Tt Can be dfned as:

=t- exp )1+ expl ).

Bome common actvation fuctons inede.


Sigmo funtn
2. tanh tuncton.
3. ReLu fuston.
4 Leahy Re lu.
5. Paraneti Re Lu.

6. saftman.
1. Sush.

Classic

You might also like