I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
6
,
No
.
2
,
J
u
n
e
2
0
1
7
,
p
p
.
74
~
78
I
SS
N:
2252
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
6
.
i2
.
p
p
7
4
-
78
74
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
A
I
Incre
m
ental
App
ro
a
ch of Neural
Netw
o
rk
in
Ba
c
k
Propa
g
a
tion
Alg
o
rith
m
s
f
o
r
W
eb Data
Mining
A.
P
.
T
a
w
da
r
,
M
.
S.
B
ewo
o
r
,
S.
H
.
P
a
t
il
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter E
n
g
in
e
e
rin
g
,
Bh
a
ra
ti
V
i
d
y
a
p
e
e
th
Un
iv
e
rsi
ty
Co
ll
e
g
e
o
f
En
g
in
e
e
rin
g
P
u
n
e
,
I
n
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
11
,
2
0
1
7
R
ev
i
s
ed
A
p
r
15
,
2
0
1
7
A
cc
ep
ted
Ma
y
24
,
2
0
1
7
T
e
x
t
Clas
si
f
ica
ti
o
n
is
a
lso
c
a
ll
e
d
a
s
Tex
t
C
a
teg
o
riza
ti
o
n
(T
C),
is
th
e
tas
k
o
f
c
las
si
fy
in
g
a
se
t
o
f
tex
t
d
o
c
u
m
e
n
ts
a
u
t
o
m
a
ti
c
a
ll
y
in
to
d
if
f
e
re
n
t
c
a
teg
o
rie
s
f
ro
m
a
p
re
d
e
f
in
e
d
se
t.
If
a
tex
t
d
o
c
u
m
e
n
t
re
late
s
to
e
x
a
c
tl
y
o
n
e
o
f
th
e
c
a
teg
o
ries
,
th
e
n
it
is
c
a
ll
e
d
a
s
sin
g
le
-
lab
e
l
c
las
sif
i
c
a
ti
o
n
tas
k
;
o
th
e
rw
ise
,
it
is
c
a
ll
e
d
a
s
m
u
lti
-
lab
e
l
c
l
a
s
sif
ic
a
ti
o
n
tas
k
.
F
o
r
In
f
o
r
m
a
ti
o
n
Re
tri
e
v
a
l
(IR)
a
n
d
M
a
c
h
in
e
L
e
a
rn
in
g
(M
L
),
T
C
u
se
s
se
v
e
r
a
l
to
o
ls
a
n
d
h
a
s
re
c
e
iv
e
d
m
u
c
h
a
tt
e
n
ti
o
n
in
t
h
e
las
t
d
e
c
a
d
e
s.
In
th
is
p
a
p
e
r,
f
irst
c
las
si
f
ies
th
e
te
x
t
d
o
c
u
m
e
n
ts
u
sin
g
M
L
P
b
a
se
d
m
a
c
h
in
e
lea
rn
in
g
a
p
p
ro
a
c
h
(B
P
P
)
a
n
d
t
h
e
n
re
tu
r
n
th
e
m
o
st
re
lev
a
n
t
d
o
c
u
m
e
n
ts.
A
n
d
a
lso
d
e
sc
rib
e
s
a
p
ro
p
o
se
d
b
a
c
k
p
ro
p
a
g
a
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
c
la
ss
i
f
ier t
h
a
t
p
e
rf
o
r
m
s c
ro
ss
v
a
li
d
a
ti
o
n
f
o
r
o
rig
in
a
l
Ne
u
ra
l
Ne
t
w
o
rk
.
In
o
r
d
e
r
to
o
p
ti
m
ize
th
e
c
las
si
f
ic
a
ti
o
n
a
c
c
u
ra
c
y
,
train
in
g
ti
m
e
.
P
ro
p
o
se
d
w
e
b
c
o
n
te
n
t
m
in
in
g
m
e
th
o
d
o
l
o
g
y
i
n
th
e
e
x
p
lo
ra
ti
o
n
w
it
h
t
h
e
a
id
o
f
BP
P
.
T
h
e
m
a
in
o
b
jec
ti
v
e
o
f
th
is
in
v
e
stig
a
ti
o
n
is
w
e
b
d
o
c
u
m
e
n
t
e
x
tra
c
ti
o
n
a
n
d
u
ti
li
z
i
n
g
d
iff
e
re
n
t
g
ro
u
p
in
g
a
lg
o
rit
h
m
.
T
h
is
w
o
rk
e
x
tri
c
a
tes
th
e
d
a
ta
f
ro
m
th
e
w
e
b
UR
L
.
K
ey
w
o
r
d
:
B
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
C
lu
s
ter
i
n
g
Featu
r
e
ex
tr
ac
tio
n
I
n
f
o
r
m
a
tio
n
e
x
tr
ac
tio
n
I
n
f
o
r
m
a
tio
n
r
etr
iev
a
l
Neu
r
al
n
et
w
o
r
k
Stea
m
i
n
g
Sto
p
w
o
r
d
Co
p
y
rig
h
t
©
2
0
1
7
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
:
A
.
P
.
T
aw
d
ar
,
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
B
h
ar
ati
Vid
y
ap
ee
th
U
n
i
v
er
s
it
y
C
o
lleg
e
o
f
E
n
g
in
ee
r
i
n
g
P
u
n
e,
I
n
d
ia
.
1.
I
NT
RO
D
UCT
I
O
N
I
n
f
o
r
m
a
tio
n
R
etr
ie
v
al
(
I
R
)
is
th
e
s
cien
ce
o
f
s
ea
r
c
h
in
g
f
o
r
in
f
o
r
m
a
tio
n
w
it
h
i
n
d
atab
ases
,
an
d
th
e
W
o
r
ld
W
id
e
W
e
b
.
T
h
e
b
r
ea
k
th
r
o
u
g
h
o
f
t
h
e
I
n
ter
n
et
a
n
d
w
e
b
s
ea
r
ch
e
n
g
i
n
e
s
h
a
v
e
u
r
g
ed
s
cien
ti
s
ts
a
n
d
lar
g
e
f
ir
m
s
to
cr
ea
te
v
er
y
lar
g
e
s
ca
le
r
etr
iev
al
s
y
s
te
m
s
to
k
ee
p
p
ac
e
w
i
th
th
e
e
x
p
o
n
e
n
tial
g
r
o
w
t
h
o
f
o
n
li
n
e
d
ata.
I
n
I
R
s
y
s
te
m
is
u
s
er
f
ir
s
t
s
u
b
m
it
s
a
q
u
er
y
w
h
ich
is
e
x
ec
u
te
d
o
v
er
th
e
r
etr
iev
al
s
y
s
t
e
m
.
T
h
e
latter
,
co
n
s
u
l
ts
a
d
atab
ase
o
f
d
o
cu
m
e
n
t
co
llecti
o
n
an
d
r
etu
r
n
s
t
h
e
m
a
tch
i
n
g
d
o
cu
m
e
n
t.
I
n
g
en
er
al,
i
n
o
r
d
er
to
co
r
r
ec
tly
clas
s
i
f
y
u
n
s
ee
n
d
o
cu
m
e
n
ts
,
it
is
n
ec
es
s
ar
y
to
tr
ain
it
w
it
h
s
o
m
e
p
r
e
-
clas
s
i
f
ied
d
o
cu
m
e
n
ts
f
r
o
m
ea
ch
ca
teg
o
r
y
t
h
at
is
tan
n
in
g
d
at
aset,
in
s
u
c
h
a
w
a
y
th
at
th
e
clas
s
i
f
ier
is
th
e
n
ab
le
to
u
n
d
er
s
tan
d
th
e
m
o
d
el
it
h
a
s
lear
n
ed
f
r
o
m
t
h
e
p
r
e
-
class
i
f
ied
d
o
cu
m
en
t
s
an
d
u
s
e
t
h
at
m
o
d
el
to
co
r
r
ec
tl
y
cla
s
s
i
f
y
t
h
e
u
n
s
ee
n
d
o
cu
m
e
n
ts
.
E
x
tr
ac
tio
n
o
f
n
et
d
ata
co
u
ld
b
e
an
i
m
p
o
r
t
an
t
m
et
h
o
d
f
o
r
d
ata
in
te
g
r
atio
n
.
W
eb
p
ag
e
s
m
a
y
g
iv
e
th
e
s
a
m
e
o
r
a
n
alo
g
o
u
s
i
n
f
o
r
m
ati
o
n
.
T
h
is
m
a
k
es
th
e
ad
d
itio
n
o
f
in
f
o
r
m
at
io
n
f
asci
n
ati
n
g
ta
s
k
.
T
h
e
d
ee
p
W
eb
co
n
ten
t
s
ar
e
ac
ce
s
s
ed
b
y
q
u
er
ies
s
u
b
m
it
ted
to
n
et
d
atab
ases
an
d
also
f
o
r
r
etr
iev
ed
d
ata.
W
eb
p
ag
es
ar
e
d
ata
r
ec
o
r
d
s
.
T
h
e
d
is
t
in
ctiv
e
W
eb
p
ag
es
ar
e
m
ad
e
p
r
o
g
r
ess
iv
el
y
a
n
d
h
ar
d
to
lis
t
b
y
r
o
u
ti
n
e
cr
a
w
ler
b
ased
w
e
b
cr
a
w
ler
s
,
i
n
p
ar
ticu
lar
Go
o
g
le
an
d
Yah
o
o
.
I
n
th
is
p
ap
er
,
d
e
p
ict
th
is
k
in
d
o
f
ex
ce
p
tio
n
al
W
eb
p
ag
es
as
d
ee
p
W
eb
p
ag
es.
A
n
o
te
w
o
r
t
h
y
i
s
s
u
e
o
f
o
n
li
n
e
w
eb
cr
a
w
ler
s
i
s
th
at
t
h
e
u
n
it
r
e
s
u
l
ts
ar
e
a
w
h
o
le
W
eb
d
o
cu
m
en
t.
Hu
m
an
e
x
er
tio
n
i
s
o
b
lig
ed
to
in
s
p
ec
t
ea
c
h
o
f
th
e
r
et
u
r
n
ed
s
ec
tio
n
s
to
s
ep
ar
ate
ex
ac
t
i
n
f
o
r
m
atio
n
.
Au
to
m
atic
in
f
o
r
m
atio
n
e
x
tr
ac
tio
n
f
r
a
m
e
w
o
r
k
s
ca
n
au
to
m
ate
th
e
tas
k
o
f
s
u
cc
es
s
f
u
l
l
y
r
ec
o
g
n
izi
n
g
t
h
e
p
er
tin
en
t
co
n
te
n
t
s
ec
tio
n
s
i
n
s
id
e
o
f
t
h
e
d
o
cu
m
en
t.
I
n
f
o
r
m
a
tio
n
E
x
tr
ac
tio
n
(
I
E
)
is
co
n
ce
r
n
ed
w
i
th
e
x
tr
ac
ti
n
g
p
er
tin
e
n
t
in
f
o
r
m
atio
n
f
r
o
m
a
g
at
h
er
in
g
o
f
ar
ch
i
v
e
s
.
I
t
in
cl
u
d
es
m
eth
o
d
s
an
d
al
g
o
r
ith
m
s
e
x
tr
a
ct
k
n
o
w
led
g
e
f
r
o
m
d
is
tin
ct
iv
e
d
ata
r
ep
o
s
ito
r
ies
s
u
ch
as
tr
an
s
ac
tio
n
a
l
d
ata
b
ases
,
d
ata
w
ar
e
h
o
u
s
e
s
,
tex
t
f
iles
,
W
W
W
an
d
th
e
n
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
n
cremen
ta
l A
p
p
r
o
a
c
h
o
f Neu
r
a
l Netw
o
r
k
in
B
a
ck
P
r
o
p
a
g
a
t
io
n
A
lg
o
r
ith
ms fo
r
… (
A
.
P
.
Ta
w
d
a
r
)
75
f
o
r
w
ar
d
s
.
Mo
s
t
o
f
t
h
e
o
n
li
n
e
r
eso
u
r
ce
s
s
q
u
ar
e
m
ea
s
u
r
e
as
m
ac
h
in
e
-
r
ea
d
ab
le
tex
t
Ma
r
k
-
u
p
L
a
n
g
u
a
g
e
(
HT
ML
)
d
o
cu
m
en
ts
,
th
at
s
q
u
ar
e
m
ea
s
u
r
e
s
ee
n
b
y
n
et
b
r
o
w
s
er
s
.
I
n
t
h
i
s
m
an
n
er
,
t
h
e
n
ee
d
f
o
r
au
to
m
a
ted
,
ad
ap
tab
le
W
eb
I
n
f
o
r
m
a
tio
n
R
etr
ie
v
al
(
I
R
)
to
o
ls
th
at
t
h
at
ex
tr
ac
t
i
n
f
o
r
m
atio
n
an
d
d
ata
f
r
o
m
th
e
o
n
lin
e
p
ag
es
an
d
tr
an
s
f
er
in
to
a
s
ig
n
i
f
ican
ce
a
n
d
v
al
u
ab
le
s
tr
u
ctu
r
e
s
f
o
r
m
o
r
e
in
v
es
tig
a
tio
n
w
il
l tu
r
n
in
to
an
ex
tr
ao
r
d
in
ar
y
n
ee
d
.
2.
RE
L
AT
E
D
WO
RK
S
T
ak
-
L
a
m
W
o
n
g
an
d
W
ai
L
a
m
[
1
]
d
ev
elo
p
ed
a
n
e
w
att
r
ib
u
te
d
is
co
v
er
y
v
ia
B
a
y
esia
n
lear
n
i
n
g
ap
p
r
o
ac
h
w
h
ich
ca
n
au
to
m
ati
ca
ll
y
ad
ap
t
t
h
e
i
n
f
o
r
m
atio
n
e
x
tr
ac
tio
n
p
atter
n
s
lear
n
ed
p
r
e
v
io
u
s
l
y
i
n
a
s
o
u
r
ce
w
eb
s
ite
to
n
e
w
u
n
s
ee
n
w
eb
s
ites
a
n
d
d
is
co
v
er
n
e
w
a
ttri
b
u
tes
to
g
eth
er
w
i
th
s
e
m
a
n
tic
lab
e
ls
.
E
x
te
n
s
iv
e
ex
p
er
i
m
e
n
ts
f
r
o
m
m
o
r
e
t
h
an
3
0
r
ea
l
tim
e
w
eb
s
ites
ar
e
in
th
r
ee
d
if
f
er
en
t
d
o
m
ai
n
s
w
er
e
co
n
d
u
cted
an
d
th
e
r
esu
lt
s
ex
h
ib
it t
h
at
th
e
f
r
a
m
e
wo
r
k
ac
h
iev
e
s
a
v
er
y
p
r
o
m
i
s
i
n
g
p
er
f
o
r
m
a
n
ce
.
R
aj
en
d
r
a
Ku
m
ar
R
o
u
l
[
2
]
h
a
s
p
r
o
p
o
s
ed
w
eb
d
o
cu
m
e
n
t
cl
u
s
ter
i
n
g
u
s
i
n
g
d
ata
m
i
n
i
n
g
.
T
h
is
p
ap
er
s
tu
d
ie
s
s
o
m
e
cl
u
s
ter
i
n
g
m
et
h
o
d
s
r
elev
an
t
to
th
e
clu
s
ter
in
g
d
o
cu
m
e
n
t
co
llectio
n
s
an
d
,
in
co
n
s
eq
u
en
ce
,
w
eb
d
ata.
T
h
is
m
eth
o
d
o
f
cl
u
s
ter
an
al
y
s
is
s
ee
m
s
to
b
e
r
elev
a
n
t
in
ap
p
r
o
ac
h
in
g
t
h
e
clu
s
ter
web
d
ata.
T
h
e
g
r
ap
h
clu
s
ter
i
n
g
is
al
s
o
d
escr
ib
ed
in
its
m
et
h
o
d
s
to
co
n
tr
ib
u
te
s
ig
n
i
f
ica
n
tl
y
in
c
lu
s
ter
in
g
web
d
ata.
B
ased
o
n
p
r
ev
io
u
s
l
y
p
r
esen
ted
in
f
o
r
m
a
t
io
n
,
th
e
co
r
e
s
ec
tio
n
p
r
o
v
id
es
an
o
v
er
v
ie
w
ap
p
r
o
ac
h
es
to
clu
s
ter
i
n
g
i
n
t
h
e
w
eb
en
v
ir
o
n
m
e
n
t.
J
ian
g
S
u
et
al
[
3
]
p
r
o
p
o
s
ed
d
ata
class
if
ica
tio
n
u
s
i
n
g
s
e
m
i
-
s
u
p
e
r
v
i
s
ed
m
u
lti
-
m
o
d
al
Nai
v
e
B
ay
e
s
.
I
t
p
r
esen
ts
Se
m
i
-
s
u
p
er
v
is
ed
F
r
eq
u
en
c
y
E
s
ti
m
ate
(
SF
E
)
,
a
n
o
v
el
s
e
m
i
-
s
u
p
er
v
i
s
ed
p
ar
a
m
et
er
lear
n
in
g
m
et
h
o
d
f
o
r
MN
B
.
T
h
e
y
f
ir
s
t
p
o
in
t
o
u
t
th
at
E
M
’
s
o
b
j
ec
tiv
e
f
u
n
ct
io
n
,
Ma
x
i
m
izi
n
g
Ma
r
g
i
n
al
L
o
g
L
i
k
eli
h
o
o
d
(
ML
L
)
,
i
s
q
u
ite
d
if
f
er
e
n
t
f
r
o
m
t
h
e
g
o
al
o
f
clas
s
i
f
icatio
n
lear
n
in
g
,
i.e
.
m
a
x
i
m
izi
n
g
co
n
d
itio
n
al
lo
g
lik
el
ih
o
o
d
(
C
L
L
)
.
T
h
en
p
r
o
p
o
s
e
SF
E
th
at
u
s
e
s
th
e
esti
m
ate
s
o
f
w
o
r
d
p
r
o
b
a
b
ilit
ies
o
b
tain
ed
f
r
o
m
u
n
lab
ell
ed
d
ata,
an
d
class
co
n
d
itio
n
al
p
r
o
b
ab
ilit
y
g
i
v
en
a
w
o
r
d
,
lear
n
ed
f
r
o
m
lab
eled
d
ata,
to
lear
n
p
ar
am
eter
s
o
f
an
MN
B
m
o
d
el.
Am
it
Ga
n
atr
a
[
4
]
h
as
p
r
o
p
o
s
ed
in
itial
clas
s
i
f
icatio
n
t
h
r
o
u
g
h
b
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
.
T
h
is
p
ap
e
r
s
a
y
s
i
n
itial
clas
s
i
f
icatio
n
u
s
i
n
g
g
en
e
tic
a
n
d
n
e
u
r
al
n
et
w
o
r
k
alg
o
r
ith
m
.
P
er
f
o
r
m
i
n
g
w
e
ig
h
t
ad
j
u
s
t
m
e
n
t
i
n
o
r
d
er
to
m
in
i
m
ize
t
h
e
Me
a
n
Sq
u
ar
e
E
r
r
o
r
b
etw
ee
n
o
b
tain
ed
o
u
t
p
u
t
an
d
d
esire
d
o
u
tp
u
t
is
t
h
e
m
ain
g
o
al
o
f
t
h
is
h
y
b
r
id
alg
o
r
ith
m
.
Fo
r
r
ed
u
cin
g
th
e
s
ea
r
c
h
s
p
ac
e
o
f
Gen
etic
alg
o
r
ith
m
it
is
b
etter
to
ap
p
l
y
b
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
f
ir
s
t.
He
n
ce
th
e
p
r
o
b
le
m
o
f
lo
ca
l
m
i
n
i
m
a
is
s
o
l
v
e
d
.
Fo
r
th
e
p
u
r
p
o
s
e
o
f
ac
ce
ler
a
tin
g
n
eu
r
al
n
et
w
o
r
k
tr
ain
i
n
g
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
ex
p
lo
it
s
t
h
e
o
p
ti
m
izatio
n
ad
v
an
ta
g
es
o
f
G
A
.
B
P
alg
o
r
ith
m
i
s
s
e
n
s
iti
v
e
to
in
itial p
ar
a
m
e
ter
s
an
d
G
A
i
s
n
o
t.
A
s
co
m
p
ar
ed
to
th
e
G
A
,
B
P
alg
o
r
ith
m
h
as
h
ig
h
co
n
v
er
g
e
n
ce
s
p
ee
d
.
Yan
L
i
u
[
5
]
p
r
o
p
o
s
ed
a
n
o
v
el
d
ee
p
lear
n
i
n
g
m
o
d
el
f
o
r
q
u
er
y
-
o
r
ien
ted
m
u
lti
d
o
cu
m
en
ts
s
u
m
m
ar
izatio
n
.
A
cc
o
r
d
in
g
l
y
,
th
e
e
m
p
ir
ical
v
a
lid
atio
n
o
n
th
r
ee
s
tan
d
ar
d
d
atasets
,
t
h
e
r
es
u
lts
n
o
t
o
n
l
y
s
h
o
w
th
e
d
i
s
ti
n
g
u
is
h
i
n
g
e
x
tr
ac
tio
n
ab
ilit
y
o
f
QODE
b
u
t
al
s
o
cle
ar
l
y
d
e
m
o
n
s
tr
ate
o
u
r
in
te
n
tio
n
to
p
r
o
v
id
e
h
u
m
an
-
lik
e
m
u
l
ti d
o
cu
m
e
n
t
s
u
m
m
ar
iz
atio
n
f
o
r
n
at
u
r
e
lan
g
u
ag
e
p
r
o
ce
s
s
i
n
g
.
Sad
u
f
,
Mo
h
d
A
r
i
f
W
an
i [
6
]
p
r
o
p
o
s
ed
th
e
co
m
p
ar
ativ
e
s
t
u
d
y
o
f
lear
n
i
n
g
i
n
n
e
u
r
al
n
et
w
o
r
k
.
C
itra
R
a
m
ad
h
e
n
al
[
7
]
h
as
p
r
o
p
o
s
ed
class
if
ica
tio
n
b
ased
o
n
er
r
o
r
r
ate.
I
n
o
r
d
er
to
m
i
n
i
m
iz
e
th
e
Me
an
Sq
u
ar
e
E
r
r
o
r
H
y
b
r
id
alg
o
r
ith
m
s
ar
e
u
s
ed
to
p
er
f
o
r
m
w
ei
g
h
t
ad
j
u
s
t
m
e
n
t.
F
ir
s
t
cr
ea
te
a
m
o
d
el
b
y
r
u
n
n
i
n
g
t
h
e
alg
o
r
ith
m
o
n
t
h
e
tr
ain
in
g
d
ata.
T
h
en
test
th
e
m
o
d
el
to
id
en
t
if
y
c
lass
o
f
n
e
w
d
ata
f
o
r
a
cl
ass
lab
el.
T
h
en
f
o
r
class
i
f
icatio
n
th
i
s
d
ata
is
g
iv
en
to
th
e
B
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
.
Af
ter
ap
p
ly
i
n
g
B
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
,
f
o
r
w
ei
g
h
t
ad
j
u
s
t
m
en
t
g
e
n
etic
alg
o
r
it
h
m
i
s
ap
p
lied
.
T
h
e
d
ev
elo
p
ed
m
o
d
el
ca
n
th
en
b
e
ap
p
lied
to
class
i
f
y
t
h
e
u
n
k
n
o
w
n
t
u
p
les
f
r
o
m
t
h
e
g
i
v
e
n
d
atab
ase
an
d
th
is
in
f
o
r
m
atio
n
m
a
y
b
e
u
s
ed
b
y
d
ec
is
io
n
m
ak
er
to
m
ak
e
u
s
e
f
u
l d
ec
is
io
n
.
B
u
t it
p
r
o
v
id
es less
e
f
f
icie
n
c
y
.
Dan
iel
So
u
d
r
y
[
8
]
p
r
o
p
o
s
ed
d
ata
class
i
f
icatio
n
i
n
n
e
u
r
al
n
et
w
o
r
k
u
s
i
n
g
d
is
cr
ete
co
n
ti
n
u
o
u
s
w
ei
g
h
t
.
I
n
in
tr
u
s
io
n
d
ete
ctio
n
an
d
cl
ass
i
f
icatio
n
u
s
in
g
b
ac
k
p
r
o
p
a
g
atio
n
n
eu
r
al
n
et
w
o
r
k
ap
p
r
o
ac
h
w
er
e
f
o
llo
w
ed
.
I
t
f
ir
s
t
co
llects
t
h
e
d
ata
s
et
th
en
t
h
e
d
ata
is
p
r
e
-
p
r
o
ce
s
s
ed
.
B
P
NN
class
if
ier
is
b
u
i
l
t
f
o
r
d
etec
tio
n
an
d
class
i
f
icatio
n
o
f
e
v
e
n
ts
.
I
n
B
P
NN
class
i
f
ier
,
f
ir
s
t
d
esi
g
n
n
et
w
o
r
k
an
d
s
e
t
p
ar
a
m
eter
s
th
en
i
n
itialize
w
ei
g
h
t
s
w
it
h
r
a
n
d
o
m
v
a
lu
e
s
;
f
in
a
ll
y
c
alcu
late
t
h
e
ac
t
u
al
o
u
tp
u
t
f
r
o
m
t
h
e
in
p
u
t.
Fi
n
all
y
,
t
h
e
R
e
s
u
lts
s
h
o
w
ed
ar
e,
it
class
i
f
ies i
n
s
tan
ce
s
in
to
s
ev
er
a
l a
ttack
t
y
p
e
s
w
it
h
lo
w
d
etec
ti
o
n
r
ate.
3.
P
RO
P
O
SE
D
SYS
T
E
M
A
w
eb
d
o
cu
m
e
n
t
i
s
s
i
m
ilar
i
n
co
n
ce
p
t
to
a
w
eb
p
ag
e.
E
v
er
y
w
eb
d
o
cu
m
en
t
h
a
s
it
s
i
n
d
iv
id
u
al
U
R
I
.
No
te
th
at
a
W
eb
d
o
cu
m
e
n
t
is
n
o
t
th
e
s
a
m
e
as
a
f
ile:
a
s
i
n
g
le
w
eb
d
o
cu
m
en
t
ca
n
b
e
ac
ce
s
s
ib
le
in
v
ar
io
u
s
ar
r
an
g
e
m
en
t
s
an
d
d
ialec
ts
,
an
d
a
s
in
g
le
d
o
cu
m
e
n
t,
f
o
r
in
s
ta
n
ce
a
P
HP
s
cr
ip
t,
m
a
y
b
e
in
c
h
ar
g
e
o
f
cr
ea
tin
g
a
s
u
b
s
ta
n
tial
n
u
m
b
er
o
f
w
eb
d
o
cu
m
e
n
t
s
w
it
h
d
if
f
er
en
t
U
R
I
s
.
A
W
eb
d
o
cu
m
en
t
i
s
ch
ar
ac
ter
ized
as
s
o
m
e
th
i
n
g
th
at
h
a
s
a
UR
I
a
n
d
ca
n
r
etu
r
n
r
ep
r
esen
tatio
n
s
o
f
th
e
id
e
n
ti
f
ie
d
ass
et
in
r
esp
o
n
s
e
o
f
HT
T
P
r
eq
u
ests
.
T
h
e
u
s
u
al
w
eb
co
n
te
n
t
e
x
tr
ac
tio
n
m
eth
o
d
s
co
n
ce
n
tr
ate
o
n
l
y
o
n
e
x
tr
ac
tin
g
t
h
e
co
n
te
n
t
w
i
th
o
u
t
ch
ec
k
i
n
g
w
h
et
h
er
th
e
co
n
te
n
t
is
r
elev
an
t
o
r
n
o
t.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
es
in
clu
d
e
d
iv
e
r
s
e
alg
o
r
ith
m
i
n
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
6
,
No
.
1
,
Ma
r
ch
2
0
1
7
:
74
–
78
76
co
m
p
ar
ati
v
e
m
an
n
er
to
ev
al
u
ate
t
h
e
p
er
f
o
r
m
a
n
ce
m
ea
s
u
r
es
o
f
t
h
e
w
eb
co
n
ten
ts
i
n
a
n
ef
f
icie
n
t
m
a
n
n
er
.
Fo
r
ex
tr
ac
tio
n
p
u
r
p
o
s
e
it o
f
f
er
s
w
eb
U
R
L
o
r
w
eb
d
o
cu
m
e
n
ts
.
I
n
p
u
t: Ge
t U
n
k
n
o
w
n
T
ex
t D
o
cu
m
e
n
t o
f
Sear
ch
ed
k
e
y
w
o
r
d
Ou
tp
u
t: L
ab
eled
to
th
e
T
ex
t D
o
cu
m
e
n
t
C
las
s
: Sp
o
r
ts
,
E
d
u
ca
tio
n
,
T
ec
h
n
o
lo
g
y
etc
Mo
d
u
les
1.
XM
L
Do
c
u
m
e
n
t E
x
tr
ac
tio
n
2.
Pa
r
s
in
g
3.
P
r
ep
r
o
ce
s
s
in
g
4.
C
las
s
i
f
icatio
n
P
ar
s
in
g
:
J
s
o
u
p
is
a
j
av
a
h
t
m
l
p
ar
s
er
is
u
s
ed
.
I
t
is
a
j
av
a
lib
r
ar
y
th
a
t
is
u
s
ed
to
p
ar
s
e
HT
M
L
d
o
cu
m
e
n
t.
J
s
o
u
p
p
r
o
v
id
es
ap
i
to
ex
tr
ac
t
an
d
m
an
ip
u
late
d
ata
f
r
o
m
U
R
L
o
r
HT
ML
f
i
le.
I
t
u
s
es
DOM
,
C
SS
a
n
d
J
q
u
er
y
-
li
k
e
m
et
h
o
d
s
f
o
r
ex
tr
ac
ti
n
g
a
n
d
m
a
n
ip
u
lat
in
g
f
ile.
T
h
e
p
ar
s
er
w
ill
m
a
k
e
ev
er
y
at
te
m
p
t
to
cr
ea
te
a
clea
n
p
ar
s
e
f
r
o
m
t
h
e
HT
ML
y
o
u
p
r
o
v
id
e,
r
eg
ar
d
les
s
o
f
w
h
et
h
er
th
e
HT
ML
i
s
w
ell
-
f
o
r
m
ed
o
r
n
o
t.
I
t h
an
d
les:
1.
U
n
clo
s
ed
tag
s
(
e.
g
.
<p
>L
o
r
e
m
<p
>I
p
s
u
m
p
ar
s
es to
<p
>L
o
r
em
</p
>
<p
>I
p
s
u
m
</p
>)
2.
I
m
p
licit ta
g
s
(
e.
g
.
a
n
a
k
ed
<td
>T
ab
le
d
ata<
/td
>
is
w
r
ap
p
ed
in
to
a
<ta
b
le>
<tr
><
td
>.
.
.
)
3.
R
eliab
l
y
cr
ea
ti
n
g
th
e
d
o
cu
m
e
n
t stru
ct
u
r
e
(
h
t
m
l c
o
n
ta
in
i
n
g
a
h
ea
d
an
d
b
o
d
y
,
an
d
o
n
l
y
ap
p
r
o
p
r
iate
elem
e
n
t
s
w
it
h
i
n
t
h
e
h
ea
d
)
T
h
e
p
ar
s
e
(
Strin
g
h
t
m
l,
Stri
n
g
b
aseUR
I
)
m
e
th
o
d
p
ar
s
es
th
e
in
p
u
t
HT
ML
i
n
to
a
n
e
w
D
o
cu
m
e
n
t.
T
h
e
b
ase
UR
I
ar
g
u
m
e
n
t
is
u
s
ed
to
r
eso
lv
e
r
elativ
e
UR
L
s
i
n
to
ab
s
o
lu
te
UR
L
s
,
an
d
s
h
o
u
l
d
b
e
s
et
to
th
e
UR
L
w
h
er
e
t
h
e
d
o
cu
m
en
t
w
as
f
etch
ed
f
r
o
m
.
I
f
t
h
at
'
s
n
o
t
ap
p
licab
l
e,
o
r
if
y
o
u
k
n
o
w
t
h
e
HT
ML
h
as
a
b
ase
ele
m
e
n
t,
y
o
u
ca
n
u
s
e
t
h
e
p
ar
s
e
(
Strin
g
h
t
m
l)
m
et
h
o
d
.
P
r
ep
r
o
ce
s
s
in
g:
it in
c
lu
d
e
t
w
o
m
o
d
u
les
:
1.
Sto
p
W
o
r
d
R
e
m
o
v
al
2.
Ste
m
m
i
n
g
Sto
p
W
o
r
d
R
e
m
o
v
al:
So
m
eti
m
es,
s
o
m
e
ex
tr
e
m
el
y
co
m
m
o
n
w
o
r
d
s
w
h
ich
w
o
u
ld
ap
p
ea
r
to
b
e
o
f
litt
le
v
alu
e
in
h
elp
i
n
g
s
elec
t
d
o
cu
m
en
ts
m
atc
h
i
n
g
a
u
s
er
n
ee
d
ar
e
ex
cl
u
d
ed
f
r
o
m
th
e
v
o
ca
b
u
lar
y
e
n
tire
l
y
.
T
h
es
e
w
o
r
d
s
ar
e
ca
lled
s
to
p
w
o
r
d
s
.
T
h
e
g
en
er
al
s
tr
ate
g
y
f
o
r
d
eter
m
in
i
n
g
a
s
to
p
lis
t
is
to
s
o
r
t
th
e
ter
m
s
b
y
co
llect
i
o
n
f
r
eq
u
e
n
c
y
,
an
d
th
en
to
ta
k
e
th
e
m
o
s
t
f
r
eq
u
en
t
ter
m
s
,
o
f
te
n
h
a
n
d
-
f
il
ter
ed
f
o
r
th
eir
s
e
m
an
t
ic
co
n
ten
t
r
elativ
e
to
th
e
d
o
m
ai
n
o
f
th
e
d
o
cu
m
e
n
t
s
b
ein
g
in
d
e
x
ed
,
as a
s
to
p
lis
t,
th
e
m
e
m
b
er
s
o
f
w
h
ic
h
ar
e
th
e
n
d
is
ca
r
d
ed
d
u
r
in
g
i
n
d
ex
i
n
g
.
Ste
m
m
i
n
g
:
I
n
t
h
is
m
e
th
o
d
w
o
r
d
s
s
h
o
r
ter
t
h
an
n
ar
e
k
ep
t
as
it
i
s
.
T
h
e
c
h
an
ce
s
o
f
o
v
er
s
te
m
m
i
n
g
i
n
cr
ea
s
es
w
h
e
n
th
e
w
o
r
d
len
g
t
h
is
s
m
all.
R
u
le
s
in
p
o
r
ter
s
te
m
m
i
n
g
al
g
o
r
ith
m
ar
e
s
ep
a
r
ated
in
to
f
iv
e
d
is
ti
n
ct
s
tep
s
:
1.
Gets r
id
o
f
p
lu
r
als a
n
d
-
ed
o
r
-
in
g
.
e
g
-
>
ca
r
ess
p
o
n
ies
-
>
p
o
n
ities
-
>
t
i c
ar
ess
-
>
ca
r
ess
ca
ts
-
>
ca
t
2.
T
u
r
n
s
ter
m
i
n
al
y
to
i
w
h
en
t
h
e
r
e
is
an
o
th
er
v
o
w
el
i
n
t
h
e
s
te
m
.
eg
h
ap
p
y
-
>h
ap
p
i
3.
Ma
p
s
d
o
u
b
le
s
u
f
f
ice
s
to
s
i
n
g
le
o
n
es.
s
o
-
izatio
n
(
=
-
ize
p
lu
s
-
atio
n
)
m
ap
s
to
-
ize
etc.
4.
Dea
ls
w
it
h
-
ic
-
,
-
f
u
ll,
-
n
e
s
s
etc
.
s
i
m
ilar
s
tr
ateg
y
to
s
tep
3
.
5.
T
ak
es o
f
f
-
an
t,
-
en
ce
etc.
3
.
1
.
I
m
ple
m
ent
a
t
io
n o
f
B
a
ck
P
ro
pa
g
a
t
io
n Alg
o
rit
hm
T
h
e
b
ac
k
-
p
r
o
p
ag
atio
n
alg
o
r
it
h
m
co
n
s
is
t
s
o
f
t
h
e
f
o
llo
w
i
n
g
s
te
p
s
:
1.
I
n
itializatio
n
:
A
t
f
ir
s
t
t
h
e
al
g
o
r
ith
m
h
as
to
b
e
in
it
ialized
co
n
s
id
er
in
g
n
o
p
r
io
r
in
f
o
r
m
ati
o
n
is
k
n
o
w
n
a
n
d
p
ick
in
g
t
h
e
s
y
n
ap
tic
w
ei
g
h
t
s
a
n
d
th
r
es
h
o
ld
s
f
r
o
m
a
u
n
i
f
o
r
m
d
is
tr
ib
u
tio
n
.
T
h
e
t
y
p
e
o
f
ac
ti
v
atio
n
f
u
n
ctio
n
is
s
ig
m
o
id
.
2.
P
r
esen
tatio
n
s
b
y
T
r
ain
in
g
E
x
a
m
p
les:
T
h
e
n
et
w
o
r
k
h
as
to
b
e
p
r
esen
ted
b
y
ep
o
ch
s
o
f
tr
ain
i
n
g
ex
a
m
p
les
to
p
er
f
o
r
m
f
o
r
w
ar
d
an
d
b
ac
k
w
ar
d
co
m
p
u
ta
tio
n
s
.
3.
Fo
r
w
ar
d
C
o
m
p
u
tatio
n
:
L
et
u
s
co
n
s
id
er
,
th
e
i
n
p
u
t
v
ec
to
r
t
o
th
e
la
y
er
o
f
s
en
s
o
r
y
n
o
d
es
is
x
(
n
)
a
n
d
t
h
e
d
esire
d
r
esp
o
n
s
e
v
ec
to
r
is
d
(
n
)
w
h
ic
h
i
s
i
n
t
h
e
o
u
tp
u
t
la
y
er
o
f
co
m
p
u
tatio
n
n
o
d
es.
I
n
f
o
r
w
ar
d
co
m
p
u
tatio
n
,
th
e
n
et
w
o
r
k
’
s
lo
ca
l
f
ield
s
an
d
f
u
n
ctio
n
s
i
g
n
als
ar
e
co
m
p
u
ted
b
y
p
r
o
ce
ed
in
g
f
o
r
w
ar
d
th
r
o
u
g
h
th
e
n
et
w
o
r
k
b
y
la
y
er
b
y
la
y
er
b
asis
.
I
m
p
le
m
e
n
tatio
n
o
f
B
ac
k
P
r
o
p
ag
atio
n
A
l
g
o
r
ith
m
t
h
e
b
ac
k
-
p
r
o
p
ag
atio
n
al
g
o
r
ith
m
c
o
n
s
i
s
ts
o
f
t
h
e
f
o
llo
w
in
g
s
tep
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
n
cremen
ta
l A
p
p
r
o
a
c
h
o
f Neu
r
a
l Netw
o
r
k
in
B
a
ck
P
r
o
p
a
g
a
t
io
n
A
lg
o
r
ith
ms fo
r
… (
A
.
P
.
Ta
w
d
a
r
)
77
1.
I
n
itializatio
n
:
A
t
f
ir
s
t
t
h
e
al
g
o
r
ith
m
h
as
to
b
e
in
it
ialized
co
n
s
id
er
in
g
n
o
p
r
io
r
in
f
o
r
m
ati
o
n
is
k
n
o
w
n
a
n
d
p
ick
in
g
t
h
e
s
y
n
ap
tic
w
ei
g
h
t
s
a
n
d
th
r
es
h
o
ld
s
f
r
o
m
a
u
n
i
f
o
r
m
d
is
tr
ib
u
tio
n
.
T
h
e
t
y
p
e
o
f
ac
ti
v
atio
n
f
u
n
ctio
n
is
s
ig
m
o
id
.
2.
P
r
esen
tatio
n
s
b
y
T
r
ain
in
g
E
x
a
m
p
les:
T
h
e
n
et
w
o
r
k
h
as
to
b
e
p
r
esen
ted
b
y
ep
o
ch
s
o
f
tr
ain
i
n
g
ex
a
m
p
les
to
p
er
f
o
r
m
f
o
r
w
ar
d
an
d
b
ac
k
w
ar
d
co
m
p
u
ta
tio
n
s
.
3.
Fo
r
w
ar
d
C
o
m
p
u
tatio
n
:
L
et
u
s
co
n
s
id
er
,
th
e
i
n
p
u
t
v
ec
to
r
t
o
th
e
la
y
er
o
f
s
en
s
o
r
y
n
o
d
es
is
x
(
n
)
a
n
d
t
h
e
d
esire
d
r
esp
o
n
s
e
v
ec
to
r
is
d
(
n
)
w
h
ic
h
i
s
i
n
t
h
e
o
u
tp
u
t
la
y
er
o
f
co
m
p
u
tatio
n
n
o
d
es.
I
n
f
o
r
w
ar
d
co
m
p
u
tatio
n
,
th
e
n
et
w
o
r
k
’
s
lo
ca
l
f
ield
s
an
d
f
u
n
ctio
n
s
i
g
n
als
ar
e
co
m
p
u
ted
b
y
p
r
o
ce
ed
in
g
f
o
r
w
ar
d
th
r
o
u
g
h
th
e
n
et
w
o
r
k
b
y
la
y
er
b
y
la
y
e
r
b
asis
.
I
f
s
ig
m
o
id
f
u
n
ctio
n
i
s
u
s
ed
,
th
e
o
u
tp
u
t
s
ig
n
al
i
s
o
b
tain
ed
b
y
t
h
e
eq
u
atio
n
b
elo
w
:
=Ҩ
j
(
V
j
(
n
)
)
(
1
)
I
f
l=1
w
h
ic
h
m
ea
n
s
t
h
e
j
n
eu
r
o
n
is
in
t
h
e
f
ir
s
t
h
id
d
en
la
y
er
t
h
en
w
e
g
et,
=X
j(
n
)
(
2
)
Her
e,
X
j(
n
)
is
t
h
e
j
th
e
le
m
e
n
t
o
f
th
e
i
n
p
u
t
v
ec
to
r
x
(
n
)
.
L
e
t,
L
i
s
t
h
e
d
ep
th
o
f
n
et
w
o
r
k
.
I
f
t
h
e
n
eu
r
o
n
j
is
in
t
h
e
o
u
tp
u
t la
y
er
t
h
at
m
ea
n
s
l=
L
th
e
n
=Oj(
n
)
(
3
)
So
th
e
er
r
o
r
s
ig
n
al
w
i
ll b
e
ej(
n
)
=d
j(
n
)
-
o
j(
n
)
(
4
)
Her
e,
d
j
(
n
)
is
th
e
j
th
ele
m
e
n
t o
f
th
e
v
ec
to
r
o
f
d
esire
d
r
esp
o
n
s
e
d
(
n
)
.
4.
I
ter
atio
n
:
Fi
n
all
y
th
e
f
o
r
w
ar
d
an
d
b
ac
k
w
ar
d
co
m
p
u
tatio
n
s
h
av
e
to
b
e
iter
ated
u
n
til
t
h
e
c
h
o
s
en
s
to
p
p
in
g
cr
iter
io
n
is
m
et.
As
th
e
n
u
m
b
er
o
f
iter
atio
n
s
i
n
cr
ea
s
es
t
h
e
m
o
m
e
n
t
u
m
a
n
d
lear
n
i
n
g
-
r
ate
p
ar
am
eter
s
ar
e
ad
j
u
s
ted
b
y
d
ec
r
ea
s
in
g
t
h
e
v
al
u
es.
4.
RE
SU
L
T
ANAL
YSI
S
T
h
is
tab
le
1
s
h
o
w
s
co
m
p
ar
ati
v
e
an
al
y
s
is
b
et
w
ee
n
B
ac
k
P
r
o
p
ag
atio
n
A
l
g
o
r
ith
m
a
n
d
Su
p
p
o
r
t
Vec
to
r
Ma
ch
i
n
e.
As s
o
o
n
as d
atase
t i
n
cr
ea
s
ed
,
th
i
s
w
ill ta
k
e
les
s
ti
m
e
th
a
n
S
VM
f
o
r
class
if
icatio
n
.
T
ab
le
1
.
C
o
m
p
ar
ativ
e
An
al
y
s
i
s
b
et
w
ee
n
B
ac
k
P
r
o
p
ag
atio
n
A
l
g
o
r
ith
m
an
d
S
u
p
p
o
r
t V
e
cto
r
Ma
ch
in
e
Fig
u
r
e
1
.
T
h
e
w
ei
g
h
t a
s
s
i
g
n
ed
f
o
r
ea
ch
Ke
y
w
o
r
d
in
Data
s
et
Fig
u
r
e
1
s
h
o
w
s
th
at
th
e
w
ei
g
h
t
as
s
ig
n
ed
f
o
r
ea
ch
k
e
y
w
o
r
d
in
d
ata
s
et.
W
h
e
n
t
h
er
e
is
n
e
w
s
ea
r
ch
i
s
h
ap
p
en
ed
th
e
n
th
at
s
ea
r
c
h
k
e
y
w
o
r
d
w
ill
m
atc
h
w
i
th
tr
ai
n
in
g
d
ataset
t
h
at
h
av
e
i
m
p
le
m
e
n
t
ed
.
W
h
en
k
e
y
w
o
r
d
n
o
t
m
atc
h
es
w
i
th
t
h
e
tr
ai
n
in
g
d
ataset
th
e
n
th
a
t k
e
y
w
o
r
d
w
ill
ca
teg
o
r
ized
as o
th
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
6
,
No
.
1
,
Ma
r
ch
2
0
1
7
:
74
–
78
78
5.
CO
NCLU
SI
O
N
First,
u
ti
lized
o
u
r
d
ev
elo
p
ed
tex
t
m
i
n
i
n
g
alg
o
r
it
h
m
s
,
in
cl
u
d
in
g
tex
t
m
i
n
i
n
g
tec
h
n
iq
u
e
s
b
ased
o
n
class
i
f
icatio
n
o
f
d
ata
i
n
s
ev
e
r
al
d
ata
co
llectio
n
s
.
Af
ter
t
h
at,
e
m
p
lo
y
e
x
iti
n
g
n
e
u
r
al
n
et
w
o
r
k
to
d
ea
l
w
ith
m
ea
s
u
r
e
th
e
tr
ai
n
i
n
g
ti
m
e
f
o
r
f
iv
e
d
ata
s
ets.
B
P
N
is
a
v
er
y
p
o
p
u
lar
alg
o
r
ith
m
in
t
h
e
ap
p
licatio
n
s
p
atter
n
m
atc
h
i
n
g
,
ch
ar
ac
ter
r
ec
o
g
n
itio
n
etc.
,
Her
e
th
is
alg
o
r
it
h
m
is
d
is
cu
s
s
ed
w
it
h
r
e
f
er
en
ce
to
t
h
e
T
ex
t c
ateg
o
r
izatio
n
p
r
o
b
le
m
.
T
h
is
al
g
o
r
ith
m
i
s
y
e
t to
b
e
i
m
p
le
m
en
ted
f
o
r
th
is
p
r
o
b
le
m
.
6.
F
UT
UR
E
WO
RK
I
m
p
le
m
e
n
tatio
n
o
f
t
h
is
alg
o
r
it
h
m
is
co
n
s
id
er
ed
as
p
ar
t
o
f
t
h
e
f
u
t
u
r
e
w
o
r
k
.
A
p
ar
t
f
r
o
m
B
P
P
th
er
e
ar
e
o
th
er
alg
o
r
ith
m
s
f
o
u
n
d
i
n
Ne
u
r
al
n
et
w
o
r
k
s
y
s
te
m
s
.
Als
o
as
in
th
is
s
y
s
te
m
s
o
m
e
f
i
x
ed
lab
elin
g
is
co
n
s
id
er
ed
in
s
tead
o
f
t
h
i
s
s
y
s
te
m
ca
n
au
to
m
atica
ll
y
g
e
n
er
ate
n
e
w
cla
s
s
a
s
n
e
w
k
e
y
w
o
r
d
is
f
o
u
n
d
.
T
r
ain
in
g
d
ataset
s
h
o
u
ld
tr
ain
as n
e
w
k
e
y
w
o
r
d
f
o
u
n
d
.
RE
F
E
R
E
NC
E
S
[1
]
T
a
k
-
La
m
W
o
n
g
a
n
d
Wai
L
a
m
,
“
L
e
a
rn
in
g
to
A
d
a
p
t
W
e
b
In
f
o
rm
a
t
io
n
Ex
trac
ti
o
n
K
n
o
w
led
g
e
a
n
d
D
isc
o
v
e
rin
g
Ne
w
A
tt
rib
u
tes
v
ia
a
Ba
y
e
sia
n
A
p
p
ro
a
c
h
”
,
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Kn
o
wled
g
e
a
n
d
Da
ta
En
g
i
n
e
e
rin
g
,
V
o
l
.
2
2
,
No
.
4
,
p
p
.
5
2
3
-
5
3
6
,
2
0
1
0
.
[2
]
R.
K.
Ro
u
l
a
n
d
S
.
K.
S
a
h
a
y
,
“
An
Ef
f
e
c
ti
v
e
A
p
p
ro
a
c
h
f
o
r
W
e
b
Do
c
u
m
e
n
t
Clas
sif
ica
ti
o
n
u
sin
g
th
e
C
o
n
c
e
p
t
o
f
A
s
so
c
iatio
n
A
n
a
l
y
sis
o
f
Da
ta
M
in
in
g
”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
Co
mp
u
ter
S
c
ien
c
e
a
n
d
E
n
g
i
n
e
e
rin
g
T
e
c
h
n
o
l
o
g
y
,
V
o
l
.
3
,
No
.
1
0
,
p
p
.
4
8
3
-
4
9
1
,
2
0
1
2
.
[3
]
Jia
n
g
S
u
,
Je
lb
e
r
S
a
y
y
a
d
S
h
irab
a
n
d
S
tan
M
a
tw
in
“
L
a
r
g
e
S
c
a
le
T
e
x
t
Clas
si
f
ic
a
ti
o
n
u
sin
g
S
e
m
i
-
su
p
e
rv
ise
d
M
u
lt
i
n
o
m
ial
Na
iv
e
Ba
y
e
s”
,
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
8
t
h
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
,
p
p
.
9
7
-
1
0
4
,
2
0
1
1
.
[4
]
Am
it
G
a
n
a
tra,
Y.
P
.
Ko
sta
,
G
a
u
ra
n
g
P
a
n
c
h
a
l
a
n
d
C
h
i
n
tan
G
a
jj
a
r,
“
In
it
ial
Clas
sif
ica
ti
o
n
T
h
ro
u
g
h
Ba
c
k
Ne
u
ra
l
Ne
tw
o
rk
F
o
ll
o
w
in
g
Op
ti
m
iz
a
ti
o
n
T
h
ro
u
g
h
GA
to
Ev
a
lu
a
te
th
e
F
it
n
e
ss
o
f
a
n
A
lg
o
rit
h
m
”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
mp
u
ter
S
c
ien
c
e
&
In
fo
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
Vo
l.
3
,
No
.
1
,
p
p
.
9
8
-
1
1
6
,
2
0
1
1
.
[5
]
Ya
n
L
iu
,
S
h
e
n
g
-
h
u
a
Z
h
o
n
g
,
W
e
n
ji
e
L
i,
“
Qu
e
ry
-
Orie
n
ted
M
u
lt
i
-
Do
c
u
m
e
n
t
S
u
m
m
a
riz
a
ti
o
n
v
ia
Un
su
p
e
rv
ise
d
De
e
p
L
e
a
rn
in
g
”
,
Pro
c
e
e
d
in
g
s o
f
th
e
T
w
e
n
ty
-
S
ixth
AA
AI
C
o
n
fer
e
n
c
e
o
n
A
rtif
icia
l
I
n
telli
g
e
n
c
e
,
p
p
.
1
6
9
9
-
1
7
0
5
,
2
0
1
2
.
[6
]
S
a
d
u
f
a
n
d
M
o
h
d
A
ri
f
W
a
n
i,
“
Co
m
p
a
ra
ti
v
e
S
tu
d
y
o
f
B
a
c
k
P
ro
p
a
g
a
ti
o
n
L
e
a
rn
in
g
A
lg
o
rit
h
m
s
f
o
r
Ne
u
ra
l
I
ss
u
e
Ne
tw
o
rk
s”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
A
d
v
a
n
c
e
d
Res
e
a
rc
h
in
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
S
o
ft
wa
re
E
n
g
i
n
e
e
rin
g
,
Vo
l.
3
,
No
.
1
2
,
p
p
.
1
1
5
1
-
1
1
5
6
,
2
0
1
3
.
[7
]
Cit
ra
Ra
m
a
d
h
e
n
a
,
A
sh
ra
f
O
s
m
a
n
Ib
ra
h
im
a
n
d
S
a
rin
a
S
u
laim
a
n
,
“
W
e
i
g
h
ts
A
d
ju
stm
e
n
t
o
f
Tw
o
-
Ter
m
Ba
c
k
-
P
r
o
p
a
g
a
ti
o
n
Ne
tw
o
rk
U
sin
g
A
d
a
p
ti
v
e
a
n
d
F
ix
e
d
L
e
a
rn
in
g
M
e
th
o
d
s”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
A
d
v
a
n
c
e
s
in
S
o
ft
Co
mp
u
t
in
g
a
n
d
it
s A
p
p
li
c
a
ti
o
n
,
Vo
l.
5
,
No
.
2
,
2
0
1
3
.
[8
]
Da
n
iel
S
o
u
d
ry
,
Ita
y
Hu
b
a
ra
a
n
d
Ro
n
M
e
ir,
“
Ex
p
e
c
tatio
n
Ba
c
k
p
ro
p
a
g
a
ti
o
n
:
P
a
ra
m
e
ter
-
F
re
e
T
r
a
in
in
g
o
f
M
u
lt
il
a
y
e
r
Ne
u
ra
l
Ne
t
w
o
rk
s
w
it
h
Co
n
ti
n
u
o
u
s
o
r
Disc
re
te
W
e
i
g
h
ts”
,
Ad
v
a
n
c
e
s
in
Ne
u
ra
l
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
,
p
p
.
9
6
3
-
9
7
1
,
2
0
1
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.