I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l
I
nte
llig
ence
(
I
J
-
AI
)
Vo
l.
4
,
No
.
2
,
J
u
n
e
201
5
,
p
p
.
6
2
~7
1
I
SS
N:
2252
-
8938
62
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
jo
u
r
n
a
l.c
o
m/o
n
lin
e/in
d
ex
.
p
h
p
/I
J
AI
Refined
Clus
terin
g
of Softw
a
re Com
po
nen
ts by
Usin
g
K
-
M
ea
n
a
nd Neural
Netw
o
rk
I
nd
u
Ver
m
a
,
A
m
a
rj
ee
t
K
a
ur
,
I
qb
a
ldeep K
a
ur
De
p
a
rt
m
e
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
,
C
h
a
n
d
ig
a
rh
U
n
iv
e
rsit
y
,
In
d
ia
.
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
6
,
2
0
1
6
R
ev
i
s
ed
A
p
r
il 9
,
2
0
1
6
A
cc
ep
ted
Ma
y
1
0
,
2
0
1
6
Da
ta
M
in
in
g
is
e
x
trac
ti
o
n
o
f
re
lev
a
n
t
in
f
o
r
m
a
ti
o
n
a
b
o
u
t
d
a
ta
se
t.
A
d
a
ta
-
w
a
r
e
h
o
u
se
is
a
lo
c
a
ti
o
n
w
h
e
re
in
f
o
rm
a
ti
o
n
is
sto
re
d
.
T
h
e
re
a
re
v
a
rio
u
s
se
rv
ice
s
o
f
d
a
ta
m
in
in
g
,
c
lu
ste
rin
g
is
o
n
e
o
f
th
e
m
.
Clu
ste
rin
g
is
a
n
e
ff
o
rt
to
g
ro
u
p
sim
il
a
r
d
a
ta
o
n
to
sin
g
le
c
lu
ste
r.
In
th
is
p
a
p
e
r
we
p
r
o
p
o
se
a
n
d
im
p
le
m
e
n
t
k
-
m
e
a
n
a
n
d
n
e
u
ra
l
n
e
tw
o
rk
f
o
r
c
lu
ste
rin
g
sa
m
e
c
o
m
p
o
n
e
n
ts
in
sin
g
le
c
lu
ste
r.
Clu
ste
rin
g
re
d
u
c
e
s
th
e
se
a
r
c
h
sp
a
c
e
b
y
g
ro
u
p
i
n
g
sim
il
a
r
tes
t
c
a
se
s
to
g
e
th
e
r
a
c
c
o
rd
in
g
to
t
h
e
re
q
u
irem
e
n
ts
a
n
d
,
h
e
n
c
e
m
in
im
izin
g
th
e
se
a
rc
h
ti
m
e
,
f
o
r
th
e
re
tri
e
v
a
l
o
f
th
e
tes
t
c
a
se
s,
re
su
lt
in
g
in
re
d
u
c
e
d
ti
m
e
c
o
m
p
lex
it
y
.
In
th
is
re
se
a
rc
h
p
a
p
e
r
w
e
p
ro
p
o
se
d
a
p
p
r
o
a
c
h
f
o
r
re
-
u
sa
b
il
it
y
o
f
tes
t
c
a
s
e
s
b
y
u
n
su
p
e
rv
ise
d
a
p
p
ro
a
c
h
a
n
d
su
p
e
rv
ise
d
a
p
p
ro
a
c
h
.
In
u
n
su
p
e
rv
ise
d
lea
rn
in
g
we
p
ro
p
o
s
e
d
k
-
m
e
a
n
a
n
d
in
su
p
e
rv
ise
d
lea
rn
in
g
n
e
u
ra
l
n
e
tw
o
rk
.
We
h
a
v
e
d
e
si
g
n
e
d
th
e
a
lg
o
rit
h
m
f
o
r
re
q
u
irem
e
n
t
a
n
d
tes
t
c
a
se
d
o
c
u
m
e
n
t
c
lu
ste
rin
g
a
c
c
o
r
d
in
g
t
o
it
s
tf
-
id
f
v
e
c
to
r
sp
a
c
e
a
n
d
th
e
o
u
tp
u
t
is
se
t
o
f
h
ig
h
ly
c
o
h
e
siv
e
p
a
tt
e
rn
g
ro
u
p
s.
K
ey
w
o
r
d
:
B
ac
k
P
r
o
p
ag
atio
n
C
lu
s
ter
i
n
g
Data
Min
i
n
g
Feed
Fo
r
w
ar
d
K
-
Me
a
n
Neu
r
al
Net
w
o
r
k
Co
p
y
rig
h
t
©
201
5
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
I
q
b
ald
ee
p
Kau
r
,
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
i
n
ee
r
in
g
,
C
h
a
n
d
ig
ar
h
U
n
iv
er
s
it
y
,
Gh
ar
u
a
n
,
I
n
d
ia.
E
m
ail:
iq
b
ald
ee
p
k
au
r
.
c
u
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
co
m
p
u
tatio
n
al
p
r
o
ce
s
s
o
f
d
eter
m
i
n
in
g
ap
p
r
o
p
r
iate
in
f
o
r
m
atio
n
o
r
p
atter
n
s
i
n
lar
g
e
d
ata
s
et
is
d
ef
in
ed
a
s
d
ata
m
in
in
g
.
C
lu
s
te
r
in
g
is
o
n
e
o
f
f
u
n
ctio
n
al
it
y
o
f
d
ata
m
i
n
i
n
g
th
at
g
r
o
u
p
ab
s
tr
ac
t
o
b
j
ec
ts
o
f
clu
s
ter
o
f
s
i
m
ilar
o
b
j
ec
ts
.
Mo
s
t
o
f
th
e
d
o
m
ai
n
s
u
s
e
cl
u
s
ter
in
g
f
o
r
th
e
r
eu
s
e
o
f
s
o
f
t
w
ar
e,
tex
t
d
o
c
u
m
e
n
t
s
,
i
m
a
g
e
s
an
d
p
atter
n
s
.
I
n
s
o
f
t
w
ar
e
e
n
g
i
n
ee
r
in
g
,
t
h
e
n
ee
d
o
f
clu
s
ter
i
n
g
ar
is
es
f
r
o
m
s
o
f
t
w
ar
e
co
m
p
o
n
en
t
c
lass
if
icat
io
n
,
co
m
p
o
n
e
n
t
cl
u
s
ter
i
n
g
,
p
er
f
o
r
m
i
n
g
s
o
f
t
w
ar
e
co
m
p
o
n
en
t
s
ea
r
ch
an
d
f
o
r
t
h
e
co
m
p
o
n
e
n
t
r
et
r
iev
al
f
r
o
m
s
o
f
t
w
ar
e
r
ep
o
s
ito
r
y
[
1
]
.
C
lu
s
ter
i
n
g
m
a
y
b
e
ca
te
g
o
r
ized
as
s
u
p
er
v
is
ed
a
n
d
u
n
s
u
p
er
v
i
s
ed
.
I
n
f
o
r
m
atio
n
r
etr
ie
v
al
a
n
d
clu
s
ter
in
g
m
ak
e
s
r
eu
s
ab
ilit
y
o
f
s
o
f
t
w
ar
e
c
o
m
p
o
n
e
n
t
s
ea
s
y
.
R
e
u
s
i
n
g
s
o
f
t
w
ar
e
is
t
h
e
p
r
o
ce
s
s
o
f
d
ev
elo
p
in
g
n
e
w
s
y
s
te
m
an
d
s
o
f
t
w
ar
e
th
at
a
lr
ea
d
y
p
r
ev
ail
r
ath
er
t
h
an
(
n
e
w
cr
ea
tio
n
)
cr
ea
tin
g
s
o
f
t
w
ar
e
s
y
s
te
m
f
r
o
m
in
itial
p
o
in
t.
T
h
er
e
ar
e
v
ar
io
u
s
cl
u
s
ter
i
n
g
tec
h
n
iq
u
es
ar
e
a
v
ailab
le
li
k
e,
De
n
s
it
y
-
b
a
s
ed
M
eth
o
d
s
,
Hier
ar
ch
ica
l
Me
th
o
d
s
,
Mo
d
el
-
b
ased
C
l
u
s
ter
i
n
g
Me
t
h
o
d
s
,
Gr
id
-
b
ased
,
P
ar
titi
o
n
i
n
g
Me
t
h
o
d
s
[
2
]
etc.
E
x
te
n
s
iv
e
r
esear
ch
h
as
b
ee
n
d
o
n
e
t
h
ese
o
v
er
lar
g
e
n
o
.
o
f
ap
p
licatio
n
s
.
T
h
er
e
is
n
o
ex
h
a
u
s
t
iv
e
r
esear
ch
u
s
in
g
a
g
g
r
e
g
atio
n
o
f
p
ar
tit
io
n
in
g
m
et
h
o
d
(
k
-
m
ea
n
)
a
n
d
Neu
r
al
Net
w
o
r
k
.
T
h
is
ca
n
f
i
n
d
its
ap
p
licatio
n
s
f
o
r
th
e
ar
ea
o
f
i
m
a
g
e
s
eg
m
e
n
tat
io
n
o
r
in
f
o
r
m
at
io
n
r
etr
iev
al.
Ho
w
ev
er
,
r
ec
en
tl
y
r
esear
ch
er
th
r
o
w
s
a
li
g
h
t
o
n
th
e
u
s
e
o
f
t
h
is
a
g
g
r
eg
ate
ap
p
r
o
ac
h
to
s
o
f
t
w
ar
e
en
g
i
n
ee
r
i
n
g
f
o
r
co
m
p
o
n
en
t
r
e
u
s
ab
ilit
y
.
T
o
r
e
d
u
ce
th
e
ti
m
e
s
p
ac
e
co
m
p
le
x
it
y
,
clu
s
ter
in
g
p
r
o
ce
s
s
o
f
s
o
f
t
w
ar
e
co
m
p
o
n
e
n
t
r
etr
iev
al
m
u
s
t b
e
a
u
to
m
ated
.
So
th
e
i
n
s
p
ir
atio
n
b
eh
in
d
th
e
d
esig
n
o
f
a
n
al
g
o
r
ith
m
i
s
a
u
to
m
at
io
n
o
f
co
m
p
o
n
e
n
t
clu
s
ter
i
n
g
.
T
h
ese
clu
s
ter
s
t
h
en
t
h
u
s
h
elp
i
n
c
h
o
o
s
in
g
t
h
e
r
eq
u
ir
ed
co
m
p
o
n
en
t
w
ith
h
ig
h
co
h
esio
n
an
d
lo
w
co
u
p
li
n
g
q
u
ic
k
l
y
a
n
d
ef
f
icien
tl
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
R
efin
ed
C
lu
s
teri
n
g
o
f S
o
ftw
a
r
e
C
o
mp
o
n
en
ts
b
y
Usi
n
g
K
-
Mea
n
a
n
d
N
eu
r
a
l Netw
o
r
k
(
I
n
d
u
V
erma
)
63
2.
AP
P
RO
ACH
In
th
e
f
o
llo
w
i
n
g
s
ec
tio
n
s
w
e
p
r
o
v
id
e
d
etails
o
n
th
e
i
n
d
iv
id
u
al
m
eth
o
d
s
,
a
n
d
th
e
n
d
escr
ib
e
h
o
w
th
e
y
w
er
e
u
s
ed
to
g
en
er
ate
o
u
r
r
es
u
lts
.
Fig
u
r
e
1
.
Me
th
o
d
o
lo
g
y
T
h
e
f
ir
s
t
s
tep
w
h
i
le
cr
ea
ti
n
g
clu
s
ter
s
o
f
d
o
cu
m
en
ts
is
to
c
o
n
v
er
t
th
e
w
h
o
le
d
o
cu
m
en
t
i
n
to
a
s
i
n
g
le
s
tate
m
en
t
.
T
h
e
s
i
m
p
lest
w
a
y
t
o
r
ep
r
esen
t
a
te
x
t
i
s
w
i
th
a
s
i
n
g
le
s
tr
i
n
g
.
T
h
ese
to
k
en
s
b
ec
o
m
e
i
n
p
u
t
f
o
r
d
ata
m
i
n
in
g
o
r
te
x
t
m
in
i
n
g
.
T
h
en
a
f
ter
t
h
at
ter
m
f
r
eq
u
e
n
c
y
(
tf
)
a
n
d
i
n
v
er
s
e
d
o
cu
m
en
t
f
r
eq
u
en
c
y
(
I
D
F
)
is
ca
lcu
lated
f
o
r
as
s
i
g
n
i
n
g
th
e
w
ei
g
h
ts
to
ea
c
h
ter
m
o
f
th
e
d
o
cu
m
e
n
t
.
T
h
e
v
al
u
es
o
f
tf
-
id
f
act
as
in
p
u
t
o
r
d
ata
p
o
in
ts
f
o
r
k
-
m
ea
n
.
2
.
1
K
-
M
ea
n
I
t
is
o
n
e
o
f
t
h
e
i
m
p
o
r
tan
t
p
ar
ts
o
f
p
ar
titi
o
n
i
n
g
m
eth
o
d
s
an
d
i
s
s
i
m
p
les
t
f
o
r
m
s
o
f
u
n
s
u
p
er
v
i
s
ed
lear
n
in
g
al
g
o
r
ith
m
s
.
T
h
is
al
g
o
r
ith
m
is
u
s
ed
f
o
r
s
o
lv
i
n
g
a
clu
s
ter
i
n
g
p
r
o
b
lem
i
n
an
y
f
ield
.
I
t
is
a
s
tr
aig
h
t
f
o
r
w
ar
d
a
n
d
e
f
f
o
r
tles
s
w
a
y
w
h
ic
h
h
elp
s
i
n
cla
s
s
i
f
y
i
n
g
a
d
ata
s
et
t
h
r
o
u
g
h
a
d
e
f
in
ite
n
u
m
b
er
o
f
c
lu
s
ter
s
(
s
a
y
c
-
cl
u
s
ter
s
)
.
I
n
itiall
y
t
h
e
n
u
m
b
er
s
o
f
cl
u
s
ter
ar
e
f
i
x
ed
s
o
th
at
it
b
ec
o
m
e
s
ea
s
y
to
cla
s
s
i
f
y
th
e
g
i
v
e
n
d
ata
s
et.
I
ts
g
o
al
is
to
p
ar
titi
o
n
k
o
b
s
er
v
atio
n
s
o
f
a
d
ata
s
et
in
t
o
c
clu
s
ter
s
,
w
h
er
e
ea
ch
o
b
s
e
r
v
atio
n
o
f
d
ata
s
et
b
elo
n
g
s
to
th
e
clu
s
ter
h
av
in
g
m
i
n
i
m
u
m
o
r
n
ea
r
est
m
ea
n
.
No
w
f
o
llo
w
t
h
e
p
r
ev
io
u
s
s
tep
f
o
r
ea
ch
p
o
in
t
b
elo
n
g
i
n
g
to
t
h
e
d
ataset
a
n
d
u
n
i
te
to
th
e
n
ea
r
est
ce
n
ter
.
T
h
e
f
ir
s
t
d
ata
s
et
i
s
co
m
p
leted
w
h
e
n
n
o
an
y
p
o
in
t
s
lef
t.
No
w
ag
a
in
ca
lcu
late
ce
n
t
r
o
id
s
as
b
ar
ce
n
ter
o
f
th
e
cl
u
s
ter
s
r
es
u
lti
n
g
f
r
o
m
p
r
ev
io
u
s
s
tep
.
Af
ter
t
h
i
s
,
t
h
e
s
a
m
e
d
ataset
p
o
in
t
s
an
d
th
e
n
ea
r
est
n
e
w
ce
n
ter
b
e
s
u
p
p
o
s
e
d
to
b
e
b
in
d
ed
to
g
eth
er
.
T
h
e
v
alu
e
o
f
k
k
ee
p
s
o
n
ch
an
g
i
n
g
u
n
til
co
n
v
er
g
en
ce
i
n
r
esu
lts
o
cc
u
r
s
.
I
n
o
u
r
w
o
r
k
th
e
t
f
-
id
f
v
al
u
es
o
f
t
h
e
in
p
u
t
d
o
cu
m
e
n
ts
w
il
l
b
ec
o
m
e
d
ata
p
o
in
ts
f
o
r
th
e
k
-
m
ea
n
al
g
o
r
it
h
m
.
T
h
e
Ob
j
ec
tiv
e
Fu
n
ctio
n
is
(
)
∑
∑
(
‖
‖
)
,
(
1
)
W
h
er
e,
‘
||
x
i
-
v
j
||
2
’
is
th
e
E
u
clid
ea
n
d
is
tan
ce
b
et
w
ee
n
x
i
a
n
d
v
j.
‘c
i
’
=
n
u
m
b
er
o
f
d
ata
p
o
in
ts
i
n
i
th
clu
s
ter
.
‘
c’
=
n
u
m
b
er
o
f
clu
s
ter
ce
n
tr
e
s
.
Alg
o
rit
h
m
f
o
r
K
-
m
ea
n:
B
E
G
I
N
X=
{X1
,
X2
,
X3
….
.
Xm
}….
S
et
o
f
d
ata
p
o
in
ts
C
=
{C1
,
C
2
,
C
3
….
.
C
n
}
….
S
et
o
f
ce
n
tr
es
I
nitia
lize
v
(
v
C
)
F
o
r
ea
ch
(
v
)
{
F
o
r
(
i=1
,
i<=
X
m
,
i++
)
{
Fin
d
Dis
tan
ce
D
i
u
s
i
n
g
E
u
clid
ea
n
d
is
ta
n
ce
If
(D
i
==
m
in
)
{
A
llo
ca
te
d
ata
p
o
in
t i
to
th
e
ce
n
ter
v
}
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
4
,
No
.
2
,
J
u
n
e
201
5
:
6
2
–
7
1
64
R
etu
r
n
D
i
E
nd
I
f
}
E
nd
F
o
r
}
(
E
nd
f
o
r
ea
ch
)
C
=
{C1
,
C
2
,
C
3
….
.
C
j
}….
Set
o
f
d
ata
p
o
in
ts
I
=
{I
1
,
I
2
,
I
3
….
.
I
n
}
….
Set o
f
ce
n
ter
s
I
n
itialize
v
(
v
I
)
F
o
r
ea
ch
(
v
)
{
F
o
r
(
k
=1
,
k
<
= C
j
,
k
++
)
{
Fin
d
Dis
tan
ce
D
k
u
s
i
n
g
E
u
clid
ea
n
d
is
ta
n
ce
If
(
Dk
co
n
v
er
g
es
w
i
th
D
i
)
T
h
en
Sto
p
E
ls
e
Go
T
o
5
}
E
nd
F
o
r
}
(
E
nd
f
o
r
ea
ch
)
E
n
d
alg
o
r
ith
m
2
.
2
.
Neura
l N
et
w
o
rk
W
h
en
A
Ne
u
r
al
Net
w
o
r
k
(
N
N)
is
o
f
ten
ca
lled
ar
ti
f
icial
n
eu
r
al
n
et
w
o
r
k
(
A
N
N)
,
is
an
esti
m
atio
n
m
et
h
o
d
o
r
m
o
d
el
w
h
ic
h
i
s
b
as
ed
o
n
n
e
u
r
o
n
s
.
I
t
p
r
o
ce
s
s
e
s
i
n
f
o
r
m
at
io
n
u
s
in
g
co
n
n
ec
t
io
n
i
s
t
ap
p
r
o
ac
h
w
h
er
e
all
th
e
n
o
d
es
ar
e
co
n
n
ec
ted
w
it
h
ea
ch
o
t
h
er
w
it
h
s
o
m
e
w
ei
g
h
t
ag
e.
Mo
s
t
l
y
a
n
e
u
r
al
n
et
w
o
r
k
i
s
a
n
ad
ap
tiv
e
s
y
s
te
m
i
n
w
h
ic
h
t
h
e
p
atter
n
c
h
an
g
es
o
n
th
e
b
asis
o
f
f
lo
w
o
f
i
n
f
o
r
m
atio
n
t
h
r
o
u
g
h
th
e
n
e
t
w
o
r
k
w
h
ile
lear
n
i
n
g
[
4
]
.
Neu
r
al
n
et
w
o
r
k
is
ab
le
to
u
s
e
s
o
m
e
h
id
d
en
u
n
k
n
o
w
n
i
n
f
o
r
m
atio
n
i
n
t
h
e
d
ata.
T
h
is
p
r
o
ce
s
s
o
f
ca
p
tu
r
i
n
g
h
id
d
en
i
n
f
o
r
m
atio
n
i
s
ca
lled
lear
n
in
g
o
r
tr
ain
i
n
g
n
et
w
o
r
k
.
I
t
is
tr
ain
ed
to
ca
r
r
y
o
u
t
s
p
ec
if
ic
f
u
n
ctio
n
b
y
m
o
d
i
f
y
i
n
g
t
h
e
v
al
u
es
o
f
t
h
e
w
ei
g
h
ts
b
et
w
ee
n
ele
m
e
n
ts
.
T
h
e
b
ac
k
p
r
o
p
ag
atio
n
n
n
is
a
m
u
lt
ila
y
er
ed
f
ee
d
f
o
r
w
ar
d
n
n
an
d
i
s
m
o
s
t
w
id
el
y
u
s
ed
f
o
r
lear
n
i
n
g
.
So
m
e
b
en
e
f
its
o
f
n
e
u
r
al
n
et
w
o
r
k
:
1.
A
d
ap
tiv
e
lear
n
i
n
g
:
A
ca
p
ac
it
y
to
ac
q
u
ir
e
s
k
i
ll
to
p
er
f
o
r
m
d
i
f
f
er
en
t
ac
ti
v
it
ies
w
h
ic
h
ar
e
d
ep
en
d
ed
o
n
th
e
p
r
o
v
id
ed
d
ata
f
o
r
tr
ain
in
g
.
2.
Self
-
Or
g
an
i
s
atio
n
:
A
n
eu
r
al
n
et
w
o
r
k
ca
n
d
e
v
elo
p
its
o
w
n
f
u
n
ct
io
n
al
b
o
d
y
t
h
at
r
etai
n
s
th
e
in
f
o
r
m
atio
n
ac
q
u
ir
ed
d
u
r
in
g
lear
n
i
n
g
p
er
io
d
.
3.
R
ea
l
T
i
m
e
Op
er
atio
n
:
Neu
r
al
Net
w
o
r
k
an
d
Data
p
r
o
ce
s
s
i
n
g
m
a
y
b
e
l
u
g
g
ed
i
n
p
ar
allel,
an
d
v
ar
io
u
s
h
ar
d
w
ar
e
d
ev
ices
ar
e
b
ein
g
m
an
u
f
ac
t
u
r
ed
an
d
d
esig
n
ed
s
p
ec
iall
y
to
ex
p
lo
it
t
h
i
s
ca
p
ab
ilit
y
i
n
t
h
e
b
est
m
an
n
er
p
o
s
s
ib
le.
Fig
u
r
e
2
.
A
r
ti
f
icial
Ne
u
r
al
Net
w
o
r
k
[
4
]
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
R
efin
ed
C
lu
s
teri
n
g
o
f S
o
ftw
a
r
e
C
o
mp
o
n
en
ts
b
y
Usi
n
g
K
-
Mea
n
a
n
d
N
eu
r
a
l Netw
o
r
k
(
I
n
d
u
V
erma
)
65
T
y
p
es o
f
Ne
u
r
al
Net
w
o
r
k
u
s
e
d
ar
e
:
1.
Feed
Fo
r
w
ar
d
A
N
N
2.
Feed
b
ac
k
ANN
2
.
3
.
F
ee
d F
o
rwa
rd
Neura
l N
et
w
o
rk
FF
n
et
w
o
r
k
is
a
s
i
m
p
le
NN
co
n
s
i
s
t
o
f
an
in
p
u
t,
o
u
tp
u
t
an
d
o
n
e
o
r
m
o
r
e
la
y
er
s
o
f
n
e
u
r
o
n
s
.
T
h
e
p
o
w
er
o
f
th
e
n
et
w
o
r
k
ca
n
b
e
n
o
ticed
b
ased
o
n
g
r
o
u
p
b
eh
av
io
r
o
f
th
e
co
n
n
ec
ted
n
e
u
r
o
n
s
a
n
d
th
e
o
u
tp
u
t
is
d
ec
id
ed
th
r
o
u
g
h
co
n
ti
n
u
o
u
s
as
s
es
s
m
e
n
t
o
f
it
s
o
u
tp
u
t
b
y
s
cr
u
ti
n
izi
n
g
.
T
h
e
v
ir
tu
e
o
f
th
i
s
n
et
w
o
r
k
is
th
at
it
b
ec
o
m
e
s
p
r
o
f
icien
t i
n
esti
m
ati
n
g
an
d
an
al
y
zi
n
g
i
n
p
u
t p
atter
n
s
.
Fig
u
r
e
3
.
Feed
Fo
r
w
ar
d
Net
w
o
r
k
[
4
]
P
r
esu
m
e
i
n
p
u
t
n
o
d
e
‘
a
’
,
h
id
d
en
n
o
d
e
‘
b
’
,
o
u
tp
u
t
n
o
d
e
‘
c
’
an
d
‘
w
’
w
ei
g
h
t
o
f
n
o
d
e.
T
h
e
in
p
u
t
la
y
er
i
s
w
h
e
n
f
ee
d
w
i
th
p
ar
ticu
lar
tr
ai
n
in
g
m
o
d
el,
th
e
s
u
m
o
f
w
e
ig
h
ted
in
p
u
t
n
o
d
es
to
th
e
b
th
n
o
d
e
in
th
e
h
id
d
en
la
y
er
is
as f
o
llo
w
s
∑
(
2
)
E
q
u
atio
n
2
is
u
s
ed
to
ca
lcu
l
ate
th
e
to
tal
in
p
u
t
to
th
e
in
p
u
t
v
al
u
es
o
f
t
h
e
in
p
u
t
la
y
er
.
T
h
e
θ
j
t
er
m
is
t
h
e
b
iased
ter
m
ad
d
ed
to
th
e
o
u
tp
u
t
v
al
u
e
o
f
th
e
i
n
p
u
t
la
y
er
t
h
at
al
w
a
y
s
h
a
s
a
v
alu
e
o
f
1
.
T
h
e
b
ias
is
ad
d
ed
t
o
r
eso
lv
e
th
e
p
r
o
b
lem
w
h
er
e
th
e
in
p
u
t
v
al
u
es
ar
e
ze
r
o
.
I
f
in
p
u
t
v
al
u
e
is
ze
r
o
,
th
en
th
e
NN
co
u
ld
n
’
t
b
e
tr
ain
ed
w
it
h
o
u
t b
ias.
T
h
e
r
esu
lti
n
g
v
alu
e
f
r
o
m
t
h
e
i
n
p
u
t
la
y
er
b
ec
o
m
es
t
h
e
in
p
u
t
f
o
r
h
id
d
en
la
y
er
.
T
h
e
p
r
o
ce
s
s
i
n
g
i
n
t
h
e
h
id
d
en
la
y
er
is
d
o
n
e
b
y
u
s
i
n
g
s
i
g
m
o
id
f
u
n
ctio
n
.
T
h
is
d
et
er
m
in
e
s
t
h
e
n
eu
r
o
n
's
o
u
tp
u
t;
a
t
y
p
ical
ac
ti
v
atio
n
f
u
n
ctio
n
u
s
ed
th
e
s
i
g
m
o
id
eq
u
atio
n
g
i
v
e
n
b
elo
w
.
(
3
)
A
b
o
v
e
eq
u
at
io
n
s
ar
e
u
s
ed
to
f
i
n
d
o
u
t th
e
o
u
tp
u
t o
f
t
h
e
o
u
tp
u
t
la
y
er
.
2
.
4
.
F
ee
d B
a
ck
o
r
B
a
ck
P
ro
pa
g
a
t
io
n Ne
ura
l N
et
w
o
rk
I
n
th
i
s
t
y
p
e
o
f
A
NN,
t
h
e
o
u
tp
u
t
is
f
ed
b
ac
k
in
to
th
e
n
et
w
o
r
k
to
i
m
p
r
o
v
is
e
t
h
e
in
p
u
t
to
ac
h
iev
e
t
h
e
b
est
p
o
s
s
ib
l
e
r
esu
lt
s
i
n
ter
n
all
y
.
T
h
e
f
ee
d
b
ac
k
n
et
w
o
r
k
f
ee
d
s
in
f
o
r
m
atio
n
b
ac
k
in
to
its
el
f
an
d
is
p
er
ti
n
en
t
i
n
s
o
lv
i
n
g
o
p
ti
m
izatio
n
p
r
o
b
lem
s
.
T
h
ese
ar
e
m
ai
n
l
y
u
s
ed
b
y
t
h
e
in
ter
n
al
s
y
s
te
m
er
r
o
r
co
r
r
ec
tio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
4
,
No
.
2
,
J
u
n
e
201
5
:
6
2
–
7
1
66
Fig
u
r
e
4
.
Feed
B
ac
k
Net
w
o
r
k
[
4
]
E
r
r
o
r
C
alcu
latio
n
s
a
n
d
W
ei
g
h
t
A
d
j
u
s
t
m
e
n
ts
ar
e
d
o
n
e
d
u
r
in
g
b
ac
k
p
r
o
p
ag
atio
n
al
s
o
ca
lled
a
s
p
r
o
p
ag
atio
n
o
f
er
r
o
r
.
I
f
th
e
ex
p
ec
ted
tar
g
et
v
al
u
e
o
f
o
u
tp
u
t o
f
n
o
d
e
c
is
t
c
a
n
d
th
e
ac
t
u
al
ac
t
iv
atio
n
v
al
u
e
o
f
t
h
e
o
u
tp
u
t n
o
d
e,
c,
is
o
c
,
th
en
t
h
e
d
if
f
er
e
n
ce
b
et
w
ee
n
t
h
e
ex
p
ec
t
ed
o
u
tp
u
t a
n
d
t
h
e
ac
t
u
al
o
u
tp
u
t is g
i
v
e
n
b
y
(
4
)
E
r
r
o
r
s
ig
n
al
at
o
u
tp
u
t n
o
d
e
c
is
:
(
)
(
)
(
5
)
W
eig
h
t a
d
j
u
s
t
m
e
n
t o
r
m
o
d
if
ic
atio
n
in
w
ei
g
h
ts
a
f
ter
ca
lcu
lati
n
g
er
r
o
r
s
ig
n
al:
(
)
(
6
)
W
h
er
e
l
r
is
lear
n
in
g
r
ate
an
d
µ
is
m
o
m
en
t
u
m
.
E
r
r
o
r
s
ig
n
al
at
h
id
d
en
n
o
d
e
b
is
:
(
)
∑
(
)
(
7
)
W
eig
h
t a
d
j
u
s
t
m
e
n
t i
n
w
eig
h
t
s
af
ter
ca
lcu
la
tin
g
er
r
o
r
s
ig
n
al:
(
)
(
8
)
W
h
er
e
is
u
p
d
ated
w
ei
g
h
t.
P
s
eudo
co
de
f
o
r
Neura
l N
et
w
o
rk
I
np
ut:
R
eq
u
ir
e
m
en
t (
R
)
T
est ca
s
es d
o
cu
m
en
t (
T
)
1.
F
o
r
I
=0
t
o
L
en
(
R
a
n
d
T
)
…
do
T
o
k
en
ize
in
p
u
t te
x
t o
f
R
an
d
T
R
e
m
o
v
e
Sto
p
w
o
r
d
Ma
k
e
a
Vec
to
r
s
p
ac
e
m
o
d
el
(
T
F
-
I
DF)
L
o
g
N/D
2.
F
o
r
I
=0
t
o
L
en
(
R
a
n
d
T
)
…
do
Vec
to
r
ize
all
Do
cu
m
e
n
ts
(
t1
,
t
2
,
t3
.
.
.
.
.
.
tn
)
f
o
r
test
ca
s
es
I
n
p
u
t t
h
ese
v
ec
to
r
izatio
n
o
n
K
-
m
ea
n
Ou
tp
u
t is K
n
u
m
b
er
o
f
clu
s
ter
Ma
k
e
K
n
u
m
b
er
o
f
cla
s
s
e
s
P
u
t f
ea
t
u
r
es o
f
t
h
ese
d
o
cu
m
en
ts
an
d
m
a
k
e
tr
ai
n
in
g
s
e
t
3.
F
o
r
I
=0
t
o
L
en
(
Do
c.
tr
ain
i
n
g
.
Featu
r
es)…
do
I
n
p
u
t o
n
Ne
u
r
al
n
et
w
o
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
R
efin
ed
C
lu
s
teri
n
g
o
f S
o
ftw
a
r
e
C
o
mp
o
n
en
ts
b
y
Usi
n
g
K
-
Mea
n
a
n
d
N
eu
r
a
l Netw
o
r
k
(
I
n
d
u
V
erma
)
67
Neu
r
al
Net
w
o
r
k
Mo
d
el
4.
F
o
r
I
=0
t
o
L
en
(
Do
c.
test
.
Feat
u
r
es)…
do
T
est
o
n
n
e
u
r
al
n
et
w
o
r
k
m
u
lti
class
i
f
ica
tio
n
m
o
d
el
u
s
e
f
o
ll
o
w
i
n
g
clu
s
ter
cla
s
s
e
s
C
2
,
C
3
,
C
4
,
C
5
,
C
6
,
an
d
C
7
Vec
to
r
ize
all
Do
cu
m
e
n
ts
(
r
1
,
r
2
,
r
3
.
.
.
.
.
.
.
.
r
n
)
f
o
r
r
eq
u
ir
em
e
n
t
s
Ana
ly
s
is
o
f
t
he
clus
t
er
o
n
1.
A
cc
u
r
ac
y
=
T
P
+
T
N/P
+N
2.
P
r
ec
is
io
n
=
T
P
/
T
P
+F
P
3.
R
ec
all
=T
P
/
T
P
+FN
O
utput
:
C
lu
s
ter
o
f
d
o
cu
m
en
t a
cc
o
r
d
in
g
to
d
o
cu
m
e
n
t si
m
ilar
it
y
3.
P
E
RF
O
RM
ANCE RES
UL
T
S AN
D
I
M
P
L
E
M
E
NT
A
T
I
O
N
T
h
is
s
ec
tio
n
d
escr
ib
es
th
e
e
x
p
er
im
e
n
tat
io
n
p
ar
t.
Fo
r
test
i
n
g
t
h
e
ef
f
ec
ti
v
e
n
es
s
o
f
t
h
e
al
g
o
r
ith
m
,
w
e
test
ed
it
o
n
r
eq
u
ir
e
m
e
n
t
a
n
d
t
est
ca
s
e
d
atase
t.
T
h
e
d
ata
s
et
is
d
iv
id
ed
r
an
d
o
m
l
y
8
0
-
2
0
p
er
ce
n
t,
o
u
t
o
f
w
h
ic
h
o
n
e
is
f
o
r
tr
ain
in
g
an
d
o
th
er
i
s
f
o
r
test
in
g
.
T
h
ese
r
esu
lts
ar
e
ca
lcu
lated
o
n
t
h
e
b
asis
o
f
co
n
f
u
s
io
n
m
atr
i
x
w
h
ich
allo
w
s
v
i
s
u
al
izin
g
t
h
e
p
er
f
o
r
m
an
ce
o
f
s
u
p
er
v
i
s
ed
alg
o
r
ith
m
.
3
.
1
.
Co
m
pa
ra
t
iv
e
Ana
ly
s
is
B
elo
w
d
is
cu
s
s
ed
r
es
u
lts
ar
e
f
o
r
clu
s
ter
s
2
,
3
,
4
,
5
,
6
,
7
.
T
ab
le
1
.
C
o
m
p
ar
ati
v
e
an
al
y
s
i
s
o
f
K
-
m
ea
n
an
d
Ne
u
r
al
Net
w
o
r
k
N
e
u
r
a
l
N
e
t
w
o
r
k
K
-
m
e
a
n
N
u
mb
e
r
o
f
C
l
u
s
t
e
r
s
A
c
c
u
r
a
c
y
P
r
e
c
i
si
o
n
R
e
c
a
l
l
A
c
c
u
r
a
c
y
P
r
e
c
i
si
o
n
R
e
c
a
l
l
Tw
o
9
6
.
3
0
%
9
6
.
1
5
%
9
6
.
6
5
%
8
1
.
6
9
%
6
6
.
6
0
%
3
1
.
7
2
%
T
h
r
e
e
9
6
.
3
0
%
9
5
.
8
3
%
9
6
.
3
0
%
7
0
.
5
4
%
5
9
.
0
0
%
4
4
.
0
6
%
F
o
u
r
9
2
.
6
0
%
6
9
.
7
3
%
6
9
.
7
2
%
7
4
.
3
1
%
5
7
.
1
0
%
5
4
.
5
2
%
F
i
v
e
9
2
.
6
0
%
5
6
.
1
2
%
5
4
.
9
2
%
8
1
.
3
4
%
5
2
.
8
0
%
6
2
.
9
4
%
S
i
x
9
6
.
3
0
%
4
8
.
4
8
%
4
8
.
4
8
%
5
4
.
3
2
%
6
1
.
9
0
%
6
0
.
8
0
%
S
e
v
e
n
8
5
.
2
0
%
4
2
.
8
6
%
3
5
.
7
1
%
5
0
.
5
9
%
5
8
.
3
0
%
7
3
.
9
5
%
3
.
2
.
Neura
l N
et
w
o
rk
Resul
t
B
elo
w
d
is
cu
s
s
ed
r
es
u
lts
ar
e
ca
lcu
lated
o
n
t
h
e
b
asis
o
f
f
e
ed
f
o
r
w
ar
d
an
d
b
ac
k
p
r
o
p
ag
atio
n
n
e
u
r
al
n
et
w
o
r
k
.
Fig
u
r
e
5
.
T
w
o
C
la
s
s
C
lass
if
ic
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
4
,
No
.
2
,
J
u
n
e
201
5
:
6
2
–
7
1
68
Fig
u
r
e
6
.
T
h
r
ee
C
lass
es C
lasi
f
icatio
n
Fig
u
r
e
7
.
Fo
u
r
C
las
s
C
la
s
s
i
f
ic
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
R
efin
ed
C
lu
s
teri
n
g
o
f S
o
ftw
a
r
e
C
o
mp
o
n
en
ts
b
y
Usi
n
g
K
-
Mea
n
a
n
d
N
eu
r
a
l Netw
o
r
k
(
I
n
d
u
V
erma
)
69
Fig
u
r
e
8
.
Fiv
e
C
lass
C
las
s
i
f
ica
tio
n
Fig
u
r
e
9
.
Six
C
lass
C
las
s
i
f
icat
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
4
,
No
.
2
,
J
u
n
e
201
5
:
6
2
–
7
1
70
Fig
u
r
e
1
0
.
Sev
en
C
las
s
C
la
s
s
i
f
icatio
n
Fig
u
r
e
1
1
.
R
ep
r
esen
ts
t
h
e
C
o
m
p
ar
i
s
o
n
o
f
Me
a
s
u
r
ed
A
cc
u
r
ac
y
o
f
K
Me
a
n
w
it
h
Me
as
u
r
ed
A
cc
u
r
ac
y
o
f
Neu
r
al
Net
w
o
r
k
.
Fig
u
r
e
1
2
.
R
ep
r
esen
ts
t
h
e
co
m
p
ar
is
o
n
o
f
m
ea
s
u
r
ed
p
r
ec
is
io
n
o
f
k
m
ea
n
w
i
th
m
ea
s
u
r
ed
p
r
ec
is
io
n
o
f
n
e
u
r
al
n
et
w
o
r
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
R
efin
ed
C
lu
s
teri
n
g
o
f S
o
ftw
a
r
e
C
o
mp
o
n
en
ts
b
y
Usi
n
g
K
-
Mea
n
a
n
d
N
eu
r
a
l Netw
o
r
k
(
I
n
d
u
V
erma
)
71
Fig
u
r
e
1
3
.
R
ep
r
esen
ts
t
h
e
C
o
m
p
ar
i
s
o
n
o
f
Me
a
s
u
r
ed
R
ec
al
l o
f
K
Me
an
w
it
h
Me
as
u
r
ed
R
e
ca
ll o
f
Neu
r
al
Net
w
o
r
k
.
4.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
w
e
h
a
v
e
r
ep
r
ese
n
ted
h
o
w
s
u
p
er
v
is
ed
an
d
u
n
s
u
p
er
v
is
ed
tec
h
n
iq
u
e
s
p
la
y
i
m
p
o
r
tan
t
r
o
le
in
co
m
p
o
n
e
n
t
b
ased
test
ca
s
e
r
ese
m
b
le
w
it
h
r
eq
u
ir
e
m
e
n
t
s
.
Un
s
u
p
er
v
is
ed
cl
u
s
ter
i
n
g
m
et
h
o
d
s
ar
e
u
s
e
to
clu
s
te
r
th
e
s
i
m
ilar
t
y
p
e
o
f
d
o
cu
m
e
n
t
s
,
r
ed
u
ce
t
h
e
s
ea
r
c
h
s
p
ac
e
a
n
d
p
r
o
v
id
e
h
i
g
h
er
e
f
f
icie
n
t a
n
d
s
u
p
er
v
is
ed
lear
n
i
n
g
i
s
u
s
e
to
a
u
to
m
ate
t
h
e
p
r
o
ce
s
s
.
I
n
th
i
s
r
esear
ch
p
ap
er
w
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
f
o
r
r
e
-
u
s
ab
ili
t
y
o
f
te
s
t
ca
s
e
s
b
y
u
n
s
u
p
er
v
is
ed
ap
p
r
o
ac
h
an
d
s
u
p
er
v
is
ed
ap
p
r
o
ac
h
.
Hen
ce
,
t
o
au
to
m
ate
t
h
e
p
r
o
ce
s
s
u
s
in
g
n
eu
r
al
n
et
w
o
r
k
i
s
ap
p
lied
o
n
th
e
d
ata
s
et
w
h
ic
h
g
a
v
e
t
h
e
clas
s
i
f
ied
d
ata
an
d
th
en
p
ar
a
m
eter
e
v
al
u
atio
n
is
d
o
n
e.
W
e
h
av
e
i
m
p
le
m
en
ted
m
u
ltil
a
y
er
n
e
u
r
a
l
n
et
w
o
r
k
w
h
e
n
t
h
er
e
ar
e
n
u
m
b
er
o
f
class
es
o
r
ca
teg
o
r
ies
ar
e
p
r
esen
t.
I
n
f
u
t
u
r
e
n
eu
r
al
n
et
w
o
r
k
ca
n
b
e
u
s
ed
w
i
th
A
d
aB
o
o
s
t le
ar
n
in
g
tec
h
n
iq
u
e
f
o
r
i
m
p
r
o
v
in
g
t
h
e
p
er
f
o
r
m
a
n
ce
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
I
w
o
u
ld
li
k
e
to
ex
p
r
ess
m
y
s
in
ce
r
e
a
n
d
d
ee
p
s
en
s
e
o
f
r
e
v
er
en
ce
f
o
r
Dr
.
Am
it
Ver
m
a
,
Hea
d
o
f
Dep
ar
t
m
en
t,
C
o
m
p
u
ter
Sc
ien
ce
an
d
E
n
g
i
n
ee
r
in
g
,
C
h
a
n
d
i
g
ar
h
Un
i
v
er
s
i
t
y
,
G
h
ar
u
an
f
o
r
h
is
g
u
id
an
ce
an
d
co
n
tin
u
o
u
s
en
co
u
r
ag
e
m
en
t.
RE
F
E
R
E
NC
E
S
[1
]
S
rin
iv
a
s,
Ch
in
tak
in
d
i,
V
a
n
g
ip
u
r
a
m
Ra
d
h
a
k
rish
n
a
,
CV
G
u
ru
Ra
o
.
Clu
ste
rin
g
a
n
d
Clas
sif
ica
ti
o
n
o
f
S
o
f
t
w
a
r
e
Co
m
p
o
n
e
n
t
f
o
r
Ef
f
ici
e
n
t
Co
m
p
o
n
e
n
t
Re
tri
e
v
a
l
a
n
d
Bu
il
d
in
g
C
o
m
p
o
n
e
n
t
Re
u
se
L
ib
ra
ries
.
Pro
c
e
d
ia
Co
m
p
u
te
r
S
c
ien
c
e
,
3
1
2
0
1
4
.
[2
]
Ro
k
a
c
h
,
L
io
r,
Od
e
d
M
a
i
m
o
n
.
Clu
ste
rin
g
m
e
th
o
d
s.
In
Da
ta
m
in
in
g
a
n
d
k
n
o
w
led
g
e
d
isc
o
v
e
r
y
h
a
n
d
b
o
o
k
.
2
0
0
5
:
3
2
1
-
3
5
2
.
S
p
rin
g
e
r
US
.
[3
]
Hu
a
n
g
,
Zh
e
x
u
e
.
Ex
ten
sio
n
s
to
th
e
k
-
m
e
a
n
s
a
lg
o
rit
h
m
f
o
r
c
lu
ste
rin
g
larg
e
d
a
ta
se
ts
w
it
h
c
a
te
g
o
rica
l
v
a
lu
e
s.
Da
ta
min
in
g
a
n
d
k
n
o
wle
d
g
e
d
isc
o
v
e
ry
2
,
1
9
9
8
;
3
:
2
8
3
-
3
0
4
.
[4
]
Ha
ss
o
u
n
,
M
o
h
a
m
a
d
H.
Fu
n
d
a
me
n
ta
ls
o
f
a
rtif
ici
a
l
n
e
u
ra
l
n
e
two
rk
s
.
P
ro
c
e
e
d
i
n
g
s o
f
th
e
IEE
E
8
4
,
1
9
9
6
;
6
:
9
0
6
.
[5
]
L
i,
Jin
h
o
n
g
,
Bi
n
g
ru
Ya
n
g
,
W
e
i
S
o
n
g
,
W
e
i
Ho
u
.
Clu
ste
ri
n
g
Fre
q
u
e
n
t
Item
se
ts
Ba
se
d
o
n
Ge
n
e
ra
to
r
s
.
In
I
n
telli
g
e
n
t
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
A
p
p
li
c
a
ti
o
n
,
2
0
0
8
.
IIT
A
'
0
8
.
S
e
c
o
n
d
In
te
rn
a
ti
o
n
a
l
S
y
m
p
o
siu
m
o
n
,
IEE
E,
2
0
0
8
;
2
:
1
0
8
3
-
1
0
8
6
.
[6
]
G
Kira
n
Ku
m
a
r,
T
Ba
la
Ch
a
r
y
,
P
P
r
e
m
c
h
a
n
d
.
A
Ne
w
a
n
d
Eff
ici
e
n
t
K
-
M
e
a
n
s
Clu
ste
rin
g
A
l
g
o
rit
h
m
.
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
A
d
v
a
n
c
e
d
Res
e
a
rc
h
in
Co
mp
u
ter
S
c
ien
c
e
a
n
d
S
o
f
twa
re
En
g
i
n
e
e
rin
g
,
2
0
1
3
;
3
(1
1
).
[7
]
F
a
ra
o
u
n
,
KM
,
A
Bo
u
k
e
li
f
.
Ne
u
ra
l
n
e
tw
o
rk
s
lea
rn
in
g
i
m
p
ro
v
e
m
e
n
t
u
sin
g
th
e
K
-
m
e
a
n
s clu
ste
rin
g
a
l
g
o
rit
h
m
to
d
e
tec
t
n
e
tw
o
rk
in
tru
sio
n
s.
I
n
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
Co
m
p
u
t
a
ti
o
n
a
l
I
n
telli
g
e
n
c
e
3
,
2
0
0
6
;
(
2
):
1
6
1
-
1
6
8
.
[8
]
A
n
e
e
th
a
,
A
S
,
S
Bo
se
.
T
h
e
c
o
m
b
in
e
d
a
p
p
r
o
a
c
h
f
o
r
a
n
o
m
a
l
y
d
e
tec
ti
o
n
u
si
n
g
n
e
u
ra
l
n
e
tw
o
rk
s
a
n
d
c
lu
ste
rin
g
tec
h
n
iq
u
e
s.
Co
m
p
u
ter
S
c
ie
n
c
e
&
En
g
i
n
e
e
rin
g
2
.
4
(2
0
1
2
):
3
7
.
[9
]
L
i,
M
in
g
,
Ha
n
g
L
i,
Zh
i
-
H
u
a
Zh
o
u
.
S
e
m
i
-
su
p
e
rv
ise
d
d
o
c
u
m
e
n
t
re
tri
e
v
a
l.
In
fo
rm
a
ti
o
n
Pro
c
e
ss
in
g
&
M
a
n
a
g
e
me
n
t
4
5
.
3
(
2
0
0
9
):
3
4
1
-
3
5
5
.
[1
0
]
Hu
a
n
g
,
Zh
e
x
u
e
.
Ex
ten
sio
n
s
to
th
e
k
-
m
e
a
n
s
a
lg
o
rit
h
m
f
o
r
c
lu
ste
rin
g
larg
e
d
a
ta
se
ts
w
it
h
c
a
te
g
o
rica
l
v
a
lu
e
s.
Da
ta
m
in
in
g
a
n
d
k
n
o
wle
d
g
e
d
isc
o
v
e
ry
2
.
3
(1
9
9
8
):
2
8
3
-
3
0
4
.
[1
1
]
S
a
k
th
i,
M
,
A
T
h
a
n
a
m
a
n
i.
A
n
En
h
a
n
c
e
d
K
M
e
a
n
s
Cl
u
ste
rin
g
u
sin
g
I
m
p
ro
v
e
d
Ho
p
f
ield
A
rti
f
icia
l
Ne
u
ra
l
Ne
tw
o
rk
a
n
d
G
e
n
e
ti
c
A
l
g
o
rit
h
m
.
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Rec
e
n
t
T
e
c
h
n
o
l
o
g
y
a
n
d
En
g
i
n
e
e
rin
g
(
IJ
RT
E)
,
2
2
7
7
-
3
8
7
8
.
[1
2
]
L
in
,
Yu
n
g
-
S
h
e
n
,
Ju
n
g
-
Yi
Jia
n
g
,
S
h
ie
-
Ju
e
L
e
e
.
A
si
m
il
a
rit
y
m
e
a
s
u
re
f
o
r
tex
t
c
las
sif
i
c
a
ti
o
n
a
n
d
c
l
u
ste
rin
g
.
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Kn
o
wled
g
e
a
n
d
Da
ta
E
n
g
in
e
e
rin
g
,
Ye
a
r
2
0
1
3
.
[1
3
]
Do
stá
l,
P
,
P
P
o
k
o
rn
ý
.
Clu
ste
r
a
n
a
ly
sis a
n
d
n
e
u
ra
l
n
e
tw
o
rk
.
Ye
a
r
2
0
0
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.