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
.
66
~
73
I
SS
N:
2252
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
6
.
i2
.
p
p
66
-
73
66
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
Deep Ma
chine
L
e
a
rning
and Neura
l Net
w
o
rk
s:
An
O
v
ervi
e
w
Cha
nd
ra
ha
s
M
is
hra
,
D.
L
.
G
u
pta
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
&
En
g
in
e
e
rin
g
,
KN
IT
S
u
lt
a
n
p
u
r
2
,
I
n
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
10
,
2
0
1
7
R
ev
i
s
ed
A
p
r
14
,
2
0
1
7
A
cc
ep
ted
Ma
y
2
3
,
2
0
1
7
De
e
p
lea
rn
in
g
is
a
tec
h
n
i
q
u
e
o
f
m
a
c
h
in
e
lea
rn
in
g
in
a
rt
if
icia
l
in
telli
g
e
n
c
e
a
re
a
.
De
e
p
lea
rn
in
g
in
a
re
f
in
e
d
"
m
a
c
h
in
e
lea
rn
in
g
"
a
lg
o
rit
h
m
th
a
t
f
a
r
su
rp
a
ss
e
s
a
c
o
n
sid
e
ra
b
le
lo
t
o
f
it
s
f
o
re
ru
n
n
e
rs
in
it
s
c
a
p
a
c
it
ies
t
o
p
e
rc
e
iv
e
s
y
ll
a
b
les
a
n
d
p
ict
u
re
.
De
e
p
le
a
rn
in
g
is
a
s
o
f
n
o
w
a
g
re
a
tl
y
d
y
n
a
m
i
c
e
x
a
m
in
a
ti
o
n
territo
ry
in
m
a
c
h
in
e
lea
rn
in
g
a
n
d
e
x
a
m
p
le
a
c
k
n
o
w
led
g
m
e
n
t
so
c
iety
.
It
h
a
s
in
c
re
a
s
e
d
c
o
lo
ss
a
l
tri
u
m
p
h
s
in
a
n
e
x
p
a
n
siv
e
z
o
n
e
o
f
u
ti
li
z
a
ti
o
n
s,
f
o
r
e
x
a
m
p
le,
sp
e
e
c
h
re
c
o
g
n
it
io
n
,
c
o
m
p
u
ter
v
isio
n
a
n
d
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
a
n
d
n
u
m
e
ro
u
s
in
d
u
stry
it
e
m
.
Ne
u
ra
l
n
e
t
w
o
rk
is
u
se
d
to
im
p
le
m
e
n
t
th
e
m
a
c
h
in
e
lea
rn
in
g
o
r
t
o
d
e
sig
n
i
n
telli
g
e
n
t
m
a
c
h
in
e
s.
In
t
h
is
p
a
p
e
r
b
rief
in
tro
d
u
c
ti
o
n
to
a
ll
m
a
c
h
in
e
lea
rn
in
g
p
a
ra
d
ig
m
a
n
d
a
p
p
li
c
a
ti
o
n
a
re
a
o
f
d
e
e
p
m
a
c
h
in
e
lea
rn
in
g
a
n
d
d
if
fe
re
n
t
t
y
p
e
s
o
f
n
e
u
ra
l
n
e
t
w
o
rk
s
w
it
h
a
p
p
li
c
a
ti
o
n
s is
d
isc
u
ss
e
d
.
K
ey
w
o
r
d
:
A
r
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
(
A
NN)
.
Au
to
m
a
tic
s
p
ee
ch
r
ec
o
g
n
itio
n
(
A
S
R
)
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
t
w
o
r
k
s
(
C
NNs)
an
d
d
ee
p
b
elief
n
et
w
o
r
k
s
(
DB
Ns)
Featu
r
e
r
ep
r
esen
tatio
n
Ma
ch
i
n
e
lear
n
i
n
g
(
M
L
)
n
eu
r
al
n
ets
m
o
d
els
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
:
C
h
a
n
d
r
ah
as M
is
h
r
a
,
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
&
E
n
g
in
ee
r
i
n
g
,
KNI
T
Su
ltan
p
u
r
2
,
I
n
d
ia.
1.
I
NT
RO
D
UCT
I
O
N
1
.
1
.
M
a
chine L
ea
rning
L
ea
r
n
i
n
g
is
a
p
r
o
ce
s
s
i
n
w
h
i
ch
as
s
o
ciatio
n
o
f
e
v
e
n
ts
w
i
th
co
n
s
eq
u
e
n
ce
s
i
s
d
o
n
e.
T
h
u
s
b
asicall
y
lear
n
in
g
i
s
a
w
a
y
to
s
u
b
s
ta
n
t
iate
th
e
ca
u
s
e
a
n
d
e
f
f
ec
t
p
r
in
cip
le.
T
h
e
s
cien
ce
o
f
d
esi
g
n
in
g
t
h
e
i
n
tell
ig
e
n
t
m
ac
h
in
e
is
r
e
f
er
r
ed
to
as
m
a
ch
in
e
lear
n
i
n
g
a
n
d
th
e
to
o
l
u
s
ed
to
d
esig
n
s
u
c
h
i
n
telli
g
e
n
t
m
ac
h
in
e
is
n
e
u
r
al
n
et
w
o
r
k
s
.
Neu
r
al
n
et
w
o
r
k
m
a
y
co
n
s
id
er
ed
as
b
lack
-
b
o
x
w
h
i
ch
g
i
v
es
s
o
m
e
d
esire
d
o
u
tp
u
t
f
o
r
th
e
g
iv
e
n
i
n
p
u
t.
I
t is a
ch
ie
v
ed
th
r
o
u
g
h
p
r
o
ce
s
s
ca
lled
tr
ain
in
g
.
I
n
co
n
tr
ast
to
m
o
s
t
co
n
v
e
n
tio
n
al
lear
n
i
n
g
m
et
h
o
d
s
,
w
h
ich
ar
e
co
n
s
id
er
ed
u
s
in
g
s
h
allo
w
-
s
tr
u
ct
u
r
ed
lear
n
in
g
ar
ch
itect
u
r
es,
d
ee
p
lear
n
in
g
r
e
f
er
s
to
m
ac
h
i
n
e
l
ea
r
n
in
g
tec
h
n
iq
u
es
th
at
u
s
e
s
u
p
er
v
i
s
ed
an
d
/o
r
u
n
s
u
p
er
v
is
ed
s
t
r
ate
g
ies
to
au
to
m
at
icall
y
lear
n
h
ier
ar
ch
ical
r
ep
r
esen
tatio
n
s
i
n
d
ee
p
ar
ch
itectu
r
e
s
f
o
r
class
i
f
icatio
n
.
I
n
s
p
ir
ed
b
y
b
io
lo
g
ical
o
b
s
er
v
atio
n
s
o
n
h
u
m
an
b
r
ai
n
m
ec
h
a
n
is
m
s
f
o
r
p
r
o
ce
s
s
in
g
o
f
n
atu
r
a
l
s
ig
n
al
s
,
d
ee
p
lear
n
i
n
g
h
as
a
ttra
cted
m
u
ch
atte
n
tio
n
f
r
o
m
t
h
e
ac
ad
e
m
ic
co
m
m
u
n
it
y
in
r
ec
en
t
y
ea
r
s
d
u
e
to
it
s
s
tate
-
of
-
th
e
-
ar
t
p
er
f
o
r
m
a
n
ce
i
n
m
a
n
y
r
esear
ch
d
o
m
ai
n
s
s
u
ch
as
s
p
ee
c
h
r
ec
o
g
n
itio
n
,
co
l
lab
o
r
ativ
e
f
ilter
i
n
g
,
an
d
co
m
p
u
ter
v
i
s
io
n
.
Dee
p
lear
n
in
g
h
a
s
ad
d
itio
n
all
y
b
ee
n
ef
f
ec
ti
v
el
y
co
n
n
ec
ted
i
n
in
d
u
s
tr
y
ite
m
s
th
a
t
ex
p
lo
it
th
e
e
x
p
an
s
i
v
e
v
o
lu
m
e
o
f
ad
v
a
n
ce
d
i
n
f
o
r
m
atio
n
.
C
o
m
p
a
n
ie
s
lik
e
Go
o
g
le,
A
p
p
le,
an
d
Fac
eb
o
o
k
,
w
h
o
co
llect
an
d
an
al
y
s
e
m
a
s
s
i
v
e
a
m
o
u
n
ts
o
f
d
ata
o
n
a
d
ail
y
b
a
s
is
,
h
av
e
b
ee
n
ag
g
r
ess
i
v
el
y
p
u
s
h
in
g
f
o
r
w
ar
d
d
ee
p
lear
n
in
g
r
elate
d
p
r
o
j
ec
ts
.
Fo
r
ex
a
m
p
le,
A
p
p
le's
Siri,
th
e
v
ir
tu
a
l
p
er
s
o
n
al
ass
is
ta
n
t
in
iP
h
o
n
es,
o
f
f
er
s
a
w
id
e
v
ar
iet
y
o
f
s
er
v
ices
i
n
cl
u
d
in
g
w
ea
t
h
er
r
ep
o
r
ts
,
s
p
o
r
t
n
ew
s
,
an
s
w
er
s
to
u
s
er
'
s
q
u
es
tio
n
s
,
an
d
r
e
m
i
n
d
er
s
etc.
b
y
u
t
ilizi
n
g
d
ee
p
lear
n
in
g
a
n
d
m
o
r
e
a
n
d
m
o
r
e
d
ata
co
llected
b
y
A
p
p
le
s
er
v
ices.
Go
o
g
le
ap
p
li
es
d
ee
p
lear
n
i
n
g
alg
o
r
it
h
m
s
to
m
a
s
s
i
v
e
c
h
u
n
k
s
o
f
m
es
s
y
d
ata
o
b
tain
ed
f
r
o
m
th
e
I
n
ter
n
et
f
o
r
Go
o
g
le's tr
an
s
lato
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
Dee
p
Ma
ch
in
e
lea
r
n
in
g
a
n
d
N
eu
r
a
l Netw
o
r
k
s
:
A
n
Ove
r
view
(
C
h
a
n
d
r
a
h
a
s
Mis
h
r
a
)
67
Dee
p
lear
n
i
n
g
r
ef
er
s
to
a
cla
s
s
o
f
M
L
tec
h
n
iq
u
e
s
,
w
h
er
e
m
an
y
la
y
er
s
o
f
in
f
o
r
m
a
tio
n
p
r
o
ce
s
s
in
g
s
tag
e
s
i
n
h
ier
ar
c
h
ical
ar
ch
i
t
ec
tu
r
es
ar
e
ex
p
lo
ited
f
o
r
u
n
s
u
p
er
v
is
ed
f
ea
t
u
r
e
lear
n
i
n
g
an
d
f
o
r
p
atter
n
class
i
f
icatio
n
.
I
t
is
i
n
th
e
in
t
er
s
ec
tio
n
s
a
m
o
n
g
t
h
e
r
esear
c
h
ar
ea
s
o
f
n
eu
r
al
n
et
w
o
r
k
,
g
r
ap
h
ical
m
o
d
elli
n
g
,
o
p
tim
izatio
n
,
p
atter
n
r
ec
o
g
n
i
t
io
n
,
an
d
s
i
g
n
al
p
r
o
ce
s
s
i
n
g
.
T
w
o
i
m
p
o
r
tan
t
r
ea
s
o
n
s
f
o
r
t
h
e
p
o
p
u
lar
it
y
o
f
d
ee
p
lear
n
in
g
to
d
a
y
ar
e
th
e
s
ig
n
i
f
i
ca
n
tl
y
lo
w
er
ed
co
s
t
o
f
co
m
p
u
tin
g
h
ar
d
w
ar
e
an
d
th
e
d
r
asti
c
all
y
i
n
cr
ea
s
ed
ch
ip
p
r
o
ce
s
s
in
g
ab
ilit
ie
s
(
e.
g
.
,
G
P
U
u
n
i
ts
)
.
Si
n
ce
2
0
0
6
,
r
esear
ch
er
s
h
av
e
d
e
m
o
n
s
tr
ated
t
h
e
s
u
cc
es
s
o
f
d
ee
p
lear
n
in
g
i
n
d
iv
er
s
e
ap
p
licatio
n
s
o
f
c
o
m
p
u
ter
v
is
io
n
,
p
h
o
n
et
ic
r
ec
o
g
n
itio
n
,
v
o
ice
s
ea
r
c
h
,
s
p
o
n
tan
eo
u
s
s
p
ee
ch
r
ec
o
g
n
itio
n
,
s
p
ee
ch
a
n
d
i
m
a
g
e
f
ea
tu
r
e
co
d
in
g
,
s
e
m
a
n
tic
u
t
ter
an
ce
class
if
ica
tio
n
,
h
an
d
-
wr
itin
g
r
ec
o
g
n
i
tio
n
,
au
d
io
p
r
o
ce
s
s
in
g
,
i
n
f
o
r
m
atio
n
r
etr
iev
al,
an
d
r
o
b
o
tics
.
B
ef
o
r
e
g
o
in
g
to
d
ea
l
w
i
t
h
d
if
f
er
en
t
m
ac
h
i
n
e
lear
n
in
g
p
ar
ad
ig
m
in
d
etail
a
b
r
ief
class
i
f
icatio
n
is
h
er
e.
W
e
u
s
e
f
o
u
r
k
e
y
at
tr
ib
u
te
to
class
i
f
y
t
h
e
m
ac
h
i
n
e
lear
n
in
g
p
ar
ad
ig
m
.
1.
2.
3.
4.
1
.
2
.
G
ener
a
t
iv
e
L
ea
rning
Gen
er
ati
v
e
lear
n
i
n
g
a
n
d
d
is
cr
i
m
i
n
ati
v
e
lear
n
i
n
g
ar
e
th
e
t
wo
m
o
s
t
p
r
ev
alen
t,
an
ta
g
o
n
i
s
tic
all
y
p
air
ed
ML
p
ar
ad
ig
m
s
d
ev
e
lo
p
ed
an
d
d
ep
lo
y
ed
in
ASR
(
Au
to
m
ati
c
s
p
ee
ch
r
ec
o
g
n
itio
n
)
.
T
h
er
e
ar
e
t
w
o
k
e
y
f
ac
to
r
s
th
at
d
i
s
ti
n
g
u
i
s
h
g
e
n
er
ati
v
e
le
ar
n
in
g
f
r
o
m
d
is
cr
i
m
i
n
ati
v
e
le
ar
n
in
g
:
th
e
n
a
tu
r
e
o
f
t
h
e
m
o
d
el
(
an
d
h
e
n
ce
th
e
d
ec
is
io
n
f
u
n
ctio
n
)
a
n
d
th
e
lo
s
s
f
u
n
ctio
n
(
i.e
.
,
th
e
co
r
e
ter
m
i
n
th
e
tr
ai
n
i
n
g
o
b
j
ec
tiv
e)
.
B
r
ief
l
y
s
p
ea
k
in
g
,
g
en
er
ati
v
e
lear
n
in
g
co
n
s
is
t
s
o
f
1.
Usi
n
g
a
g
e
n
er
ati
v
e
m
o
d
el,
an
d
2.
A
d
o
p
tin
g
a
tr
ain
i
n
g
o
b
j
ec
tiv
e
f
u
n
ctio
n
b
ased
o
n
th
e
j
o
in
t li
k
elih
o
o
d
lo
s
s
d
ef
i
n
ed
o
n
th
e
g
e
n
er
ativ
e
m
o
d
el.
Dis
cr
i
m
in
at
iv
e
lear
n
in
g
,
o
n
t
h
e
o
th
er
h
an
d
,
r
eq
u
ir
es e
it
h
er
1.
Usi
n
g
a
d
is
cr
i
m
i
n
ati
v
e
m
o
d
el,
o
r
2.
A
p
p
l
y
in
g
a
d
is
cr
i
m
in
at
iv
e
tr
ai
n
in
g
o
b
j
ec
tiv
e
f
u
n
c
tio
n
to
a
g
e
n
er
ativ
e
m
o
d
el.
I
n
th
i
s
an
d
th
e
n
ex
t
s
ec
tio
n
s
,
w
e
w
il
l
d
is
cu
s
s
g
e
n
er
ati
v
e
v
s
.
d
is
cr
i
m
in
a
tiv
e
lear
n
in
g
f
r
o
m
b
o
th
th
e
m
o
d
el
a
n
d
lo
s
s
f
u
n
ctio
n
p
er
s
p
ec
tiv
es.
W
h
i
le
h
i
s
to
r
icall
y
t
h
e
r
e
h
as
b
ee
n
a
s
tr
o
n
g
a
s
s
o
ciatio
n
b
et
w
ee
n
a
m
o
d
el
an
d
th
e
lo
s
s
f
u
n
ctio
n
c
h
o
s
en
t
o
tr
ain
t
h
e
m
o
d
el,
t
h
er
e
h
a
s
b
ee
n
n
o
n
ec
e
s
s
ar
y
p
air
in
g
o
f
t
h
ese
t
w
o
co
m
p
o
n
en
t
in
th
e
l
iter
atu
r
e
1
.
3
.
Dis
cr
e
m
i
na
t
iv
e
L
ea
rning
As
d
is
c
u
s
s
ed
ea
r
lier
,
th
e
p
ar
a
d
ig
m
o
f
d
is
cr
i
m
i
n
ati
v
e
lear
n
i
n
g
in
v
o
lv
e
s
eit
h
er
u
s
i
n
g
a
d
is
cr
i
m
i
n
ati
v
e
m
o
d
el
o
r
ap
p
l
y
i
n
g
d
i
s
cr
i
m
i
n
a
tiv
e
tr
ai
n
in
g
to
a
g
en
er
ati
v
e
m
o
d
el.
I
n
th
is
s
ec
t
io
n
,
w
e
f
ir
s
t
p
r
o
v
id
e
a
g
en
er
al
d
is
cu
s
s
io
n
o
f
t
h
e
d
is
cr
i
m
in
a
ti
v
e
m
o
d
els
an
d
o
f
t
h
e
d
is
cr
i
m
i
n
ati
v
e
lo
s
s
f
u
n
ctio
n
s
u
s
ed
in
t
r
ain
in
g
,
f
o
llo
w
ed
b
y
an
o
v
er
v
ie
w
o
f
t
h
e
u
s
e
o
f
d
is
cr
i
m
i
n
ati
v
e
lear
n
in
g
i
n
AS
R
ap
p
licatio
n
s
i
n
cl
u
d
in
g
i
ts
s
u
c
ce
s
s
f
u
l
h
y
b
r
id
w
it
h
g
en
er
ati
v
e
lear
n
in
g
.
M
o
dels
:
Dis
cr
i
m
in
at
iv
e
m
o
d
el
s
m
a
k
e
d
ir
ec
t
u
s
e
o
f
th
e
co
n
d
itio
n
al
r
elatio
n
o
f
lab
els
g
iv
e
n
in
p
u
t
v
ec
to
r
s
.
On
e
m
aj
o
r
s
ch
o
o
l
o
f
s
u
ch
m
o
d
els
ar
e
r
ef
er
r
e
d
to
as
B
a
ye
s
ia
n
Min
imu
m
R
is
k
(
B
M
R
)
c
lass
i
f
ier
s
.
S
h
o
w
n
i
n
eq
u
atio
n
1
.
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
:
66
–
73
68
(
1
)
L
o
s
s
F
u
nct
io
ns
:
T
h
is
s
ec
tio
n
i
n
tr
o
d
u
ce
s
a
n
u
m
b
er
o
f
d
is
cr
i
m
i
n
ati
v
e
lo
s
s
f
u
n
ctio
n
s
.
T
h
e
f
ir
s
t
g
r
o
u
p
o
f
lo
s
s
f
u
n
ctio
n
s
ar
e
b
ased
o
n
p
r
o
b
ab
ilis
tic
m
o
d
els,
w
h
ile
t
h
e
s
ec
o
n
d
g
r
o
u
p
o
n
th
e
n
o
tio
n
o
f
ma
r
g
in
.
1
)
P
r
o
b
a
b
ilit
y
-
B
a
s
ed
Lo
s
s
:
Si
m
i
lar
to
t
h
e
j
o
in
t
li
k
eli
h
o
o
d
lo
s
s
d
is
c
u
s
s
e
d
in
th
e
p
r
ec
ed
in
g
s
ec
tio
n
o
n
g
en
er
ati
v
e
lear
n
i
n
g
,
co
n
d
itio
n
a
l
likelih
o
o
d
lo
s
s
is
a
p
r
o
b
ab
i
lit
y
-
b
ased
lo
s
s
f
u
n
ctio
n
b
u
t
i
s
d
ef
in
ed
u
p
o
n
th
e
co
n
d
itio
n
al
r
elatio
n
o
f
cla
s
s
la
b
els g
i
v
en
i
n
p
u
t
f
ea
t
u
r
es
.
S
h
o
w
n
i
n
eq
u
atio
n
2
:
(
2
)
T
h
is
lo
s
s
f
u
n
ctio
n
is
s
tr
o
n
g
l
y
tied
to
p
r
o
b
ab
ilis
tic
d
is
cr
i
m
i
n
ati
v
e
m
o
d
els
s
u
ch
as
co
n
d
it
io
n
al
lo
g
lin
ea
r
m
o
d
el
s
a
n
d
ML
P
s
,
w
h
ile
th
e
y
ca
n
b
e
ap
p
lied
to
g
e
n
er
ativ
e
m
o
d
els
as
w
e
ll,
lead
in
g
to
a
s
c
h
o
o
l
o
f
d
is
cr
i
m
i
n
ati
v
e
tr
ai
n
in
g
m
et
h
o
d
s
w
h
ic
h
w
ill
b
e
d
is
cu
s
s
ed
s
h
o
r
tly
.
Mo
r
eo
v
er
,
co
n
d
itio
n
al
li
k
eli
h
o
o
d
lo
s
s
ca
n
b
e
n
atu
r
al
l
y
e
x
ten
d
ed
to
p
r
ed
icti
n
g
s
tr
u
ct
u
r
e
o
u
tp
u
t.
Fo
r
ex
a
m
p
le,
w
h
e
n
ap
p
l
y
in
g
to
Ma
r
k
o
v
r
an
d
o
m
f
ie
ld
s
,
w
e
o
b
tain
th
e
tr
ai
n
in
g
o
b
j
ec
tiv
e
o
f
co
n
d
itio
n
a
l r
an
d
o
m
f
ield
s
(
C
R
Fs
)
: b
y
eq
u
atio
n
3
(
3
)
No
te
th
at
in
m
o
s
t
o
f
th
e
M
L
as
w
ell
a
s
th
e
A
S
R
liter
atu
r
e,
o
n
e
o
f
ten
ca
ll
s
th
e
tr
ain
in
g
m
e
t
h
o
d
u
s
i
n
g
th
e
co
n
d
itio
n
a
l lik
e
lih
o
o
d
lo
s
s
ab
o
v
e
as si
m
p
l
y
m
a
x
i
m
al
li
k
e
lih
o
o
d
esti
m
a
tio
n
(
M
L
E
)
.
A
g
e
n
er
aliza
tio
n
o
f
co
n
d
itio
n
al
lik
e
lih
o
o
d
lo
s
s
is
Mi
n
i
m
u
m
B
a
y
e
s
R
is
k
tr
ai
n
i
n
g
.
T
h
is
i
s
co
n
s
i
s
te
n
t
w
it
h
th
e
cr
i
ter
io
n
o
f
MB
R
cl
ass
i
f
ier
s
d
escr
ib
ed
i
n
t
h
e
p
r
ev
io
u
s
s
u
b
s
ec
tio
n
.
T
h
e
lo
s
s
f
u
n
ctio
n
o
f
(
MB
R
)
i
n
tr
ain
i
n
g
i
s
g
i
v
e
n
b
y
eq
u
atio
n
4
(
4
)
1
.
4
.
Se
m
i
-
Su
perv
is
ed
a
n
d
Act
iv
e
L
ea
rning
T
h
e
p
r
ec
ed
in
g
o
v
er
v
ie
w
o
f
g
en
er
ativ
e
an
d
d
is
cr
i
m
i
n
ati
v
e
ML
p
ar
ad
ig
m
s
u
s
es
t
h
e
at
tr
ib
u
tes
o
f
lo
s
s
an
d
d
ec
is
io
n
f
u
n
c
tio
n
s
to
o
r
g
an
ize
a
m
u
ltit
u
d
e
o
f
ML
tec
h
n
iq
u
e
s
.
I
n
th
is
s
ec
tio
n
,
w
e
u
s
e
a
d
if
f
er
en
t
s
et
o
f
attr
ib
u
tes,
n
a
m
el
y
t
h
e
n
at
u
r
e
o
f
th
e
tr
ain
i
n
g
d
ata
in
r
elatio
n
to
th
eir
class
lab
els.
Dep
en
d
in
g
o
n
t
h
e
w
a
y
th
a
t
tr
ain
i
n
g
s
a
m
p
le
s
ar
e
lab
elled
o
r
o
th
er
w
is
e,
w
e
ca
n
cla
s
s
i
f
y
m
an
y
ex
i
s
ti
n
g
M
L
tec
h
n
iq
u
es
in
to
s
ev
er
al
s
ep
ar
ate
p
ar
ad
ig
m
s
,
m
o
s
t
o
f
w
h
ic
h
h
a
v
e
b
ee
n
i
n
u
s
e
i
n
th
e
AS
R
p
r
ac
tice.
Su
p
er
v
i
s
ed
lear
n
in
g
a
s
s
u
m
es
th
at
a
ll
tr
ai
n
i
n
g
s
a
m
p
les
ar
e
lab
elled
,
w
h
ile
u
n
s
u
p
er
v
is
ed
lear
n
in
g
a
s
s
u
m
es
none
.
Se
m
i
-
s
u
p
er
v
i
s
ed
lea
r
n
in
g
,
as
t
h
e
n
a
m
e
s
u
g
g
e
s
ts
,
as
s
u
m
e
s
th
at
b
o
th
la
b
elled
an
d
u
n
lab
elled
tr
ain
in
g
s
a
m
p
les
ar
e
av
ailab
le.
Su
p
er
v
is
ed
,
u
n
s
u
p
er
v
is
ed
an
d
s
e
m
i
-
s
u
p
er
v
i
s
ed
l
ea
r
n
in
g
ar
e
ty
p
icall
y
r
e
f
er
r
ed
to
u
n
d
er
th
e
p
a
s
s
ive
lea
r
n
in
g
s
ettin
g
,
w
h
er
e
lab
elled
tr
ain
i
n
g
s
a
m
p
les
ar
e
g
e
n
er
ated
r
an
d
o
m
l
y
ac
co
r
d
in
g
to
an
u
n
k
n
o
w
n
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
.
I
n
co
n
tr
ast,
a
ctive
lea
r
n
in
g
is
a
s
ettin
g
w
h
er
e
th
e
lear
n
er
ca
n
in
tel
lig
e
n
tl
y
ch
o
o
s
e
w
h
ic
h
s
a
m
p
les
to
lab
el,
w
h
ich
w
e
w
ill
d
is
c
u
s
s
at
th
e
en
d
o
f
t
h
is
s
ec
tio
n
.
I
n
th
is
s
ec
tio
n
,
w
e
co
n
ce
n
tr
ate
m
ain
l
y
o
n
s
e
m
i
-
s
u
p
er
v
is
ed
an
d
ac
tiv
e
lear
n
in
g
p
ar
ad
ig
m
s
.
T
h
is
is
b
ec
au
s
e
s
u
p
er
v
is
ed
lear
n
in
g
is
r
ea
s
o
n
ab
l
y
w
e
ll
u
n
d
er
s
to
o
d
an
d
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
d
o
es
n
o
t
d
ir
ec
tl
y
ai
m
at
p
r
ed
ictin
g
o
u
tp
u
ts
f
r
o
m
i
n
p
u
t
s
(
an
d
h
en
ce
is
b
e
y
o
n
d
th
e
f
o
c
u
s
o
f
th
is
ar
ticle)
.
W
e
w
il
l
co
v
er
th
ese
t
w
o
to
p
ics o
n
l
y
b
r
ief
l
y
.
1
.
4
.
1
.
Su
perv
is
e
d L
ea
rni
ng
I
n
s
u
p
er
v
is
ed
lear
n
i
n
g
,
th
e
tr
ain
i
n
g
s
et
co
n
s
is
t
s
o
f
p
air
s
o
f
in
p
u
t
s
an
d
o
u
tp
u
ts
d
r
a
w
n
f
r
o
m
a
j
o
in
t
d
is
tr
ib
u
tio
n
.
U
s
in
g
n
o
tatio
n
s
i
n
tr
o
d
u
ce
d
b
y
eq
u
atio
n
5
:
(
5
)
T
h
e
lear
n
in
g
o
b
j
ec
tiv
e
i
s
a
g
ai
n
e
m
p
ir
ical
r
is
k
m
i
n
i
m
izatio
n
w
it
h
r
e
g
u
lar
izatio
n
,
i.e
.
w
h
er
e
b
o
th
in
p
u
t
d
ata
an
d
th
e
co
r
r
esp
o
n
d
in
g
o
u
tp
u
t
lab
el
s
ar
e
p
r
o
v
id
ed
.
No
tice
th
at
t
h
er
e
m
a
y
e
x
i
s
t
m
u
lt
ip
le
lev
els
o
f
lab
e
l
v
ar
iab
les,
n
o
tab
l
y
i
n
A
S
R
.
I
n
t
h
is
ca
s
e,
w
e
s
h
o
u
ld
d
is
ti
n
g
u
is
h
b
et
w
ee
n
t
h
e
f
u
ll
y
s
u
p
er
v
is
ed
ca
s
e,
w
h
er
e
lab
els
o
f
all
le
v
el
s
ar
e
k
n
o
w
n
,
t
h
e
p
ar
tiall
y
s
u
p
er
v
is
ed
ca
s
e,
w
h
er
e
lab
els
at
ce
r
tai
n
le
v
el
s
ar
e
m
is
s
i
n
g
.
I
n
A
SR
,
f
o
r
ex
a
m
p
le,
it
is
o
f
ten
t
h
e
ca
s
e
t
h
at
t
h
e
tr
ain
i
n
g
s
et
co
n
s
i
s
ts
o
f
w
a
v
ef
o
r
m
s
a
n
d
th
eir
co
r
r
es
p
o
n
d
in
g
w
o
r
d
-
le
v
el
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
Dee
p
Ma
ch
in
e
lea
r
n
in
g
a
n
d
N
eu
r
a
l Netw
o
r
k
s
:
A
n
Ove
r
view
(
C
h
a
n
d
r
a
h
a
s
Mis
h
r
a
)
69
tr
an
s
cr
ip
tio
n
s
as
t
h
e
lab
el
s
,
wh
ile
t
h
e
p
h
o
n
e
-
lev
el
tr
an
s
cr
ip
t
io
n
s
a
n
d
ti
m
e
ali
g
n
m
e
n
t
i
n
f
o
r
m
atio
n
b
et
w
ee
n
t
h
e
w
a
v
e
f
o
r
m
s
a
n
d
th
e
co
r
r
esp
o
n
d
i
n
g
p
h
o
n
es a
r
e
m
i
s
s
in
g
.
1
.
4
.
2
.
Uns
u
perv
is
ed
L
ea
rning
I
n
ML
,
u
n
s
u
p
er
v
i
s
ed
lear
n
i
n
g
in
g
e
n
er
al
r
ef
er
s
to
lear
n
i
n
g
w
it
h
t
h
e
in
p
u
t
d
ata
o
n
l
y
.
T
h
is
lear
n
i
n
g
p
ar
ad
ig
m
o
f
te
n
ai
m
s
at
b
u
ild
in
g
r
ep
r
esen
ta
tio
n
s
o
f
t
h
e
in
p
u
t
th
at
ca
n
b
e
u
s
ed
f
o
r
p
r
ed
ictio
n
,
d
ec
is
io
n
m
a
k
in
g
o
r
class
i
f
icatio
n
,
a
n
d
d
ata
co
m
p
r
e
s
s
io
n
.
Fo
r
ex
a
m
p
le,
d
en
s
it
y
esti
m
atio
n
,
cl
u
s
ter
i
n
g
,
p
r
in
cip
le
co
m
p
o
n
e
n
t
an
al
y
s
is
a
n
d
in
d
ep
en
d
e
n
t
co
m
p
o
n
en
t
a
n
al
y
s
i
s
ar
e
all
i
m
p
o
r
t
an
t
f
o
r
m
s
o
f
u
n
s
u
p
er
v
i
s
ed
lear
n
in
g
.
Use
o
f
v
ec
to
r
q
u
an
tizat
io
n
(
VQ)
to
p
r
o
v
id
e
d
is
cr
ete
in
p
u
ts
to
A
S
R
is
o
n
e
ea
r
l
y
s
u
cc
es
s
f
u
l
ap
p
licati
o
n
o
f
u
n
s
u
p
er
v
is
ed
lear
n
in
g
to
A
S
R
[
8
]
.
Mo
r
e
r
ec
en
tl
y
,
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
h
a
s
b
ee
n
d
ev
elo
p
ed
as
a
co
m
p
o
n
en
t
o
f
s
ta
g
ed
h
y
b
r
id
g
e
n
er
ati
v
e
-
d
i
s
cr
i
m
i
n
at
iv
e
p
ar
ad
ig
m
i
n
M
L
.
T
h
is
e
m
er
g
i
n
g
tech
n
iq
u
e,
b
ased
o
n
th
e
d
ee
p
lear
n
in
g
f
r
a
m
e
w
o
r
k
,
is
b
eg
in
n
i
n
g
to
m
ak
e
i
m
p
ac
t
o
n
A
S
R
.
L
ea
r
n
in
g
s
p
ar
s
e
s
p
ee
ch
r
ep
r
esen
tatio
n
s
,
to
b
e
talk
ed
ab
o
u
t
ca
n
lik
e
w
is
e
b
e
v
ie
w
ed
as
u
n
s
u
p
er
v
is
ed
f
ea
tu
r
e
lear
n
in
g
o
r
lear
n
in
g
f
ea
t
u
r
e
r
ep
r
esen
ta
tio
n
s
i
n
ab
s
en
ce
o
f
class
i
f
icatio
n
lab
els.
1
.
4
.
3
.
Se
m
i
-
S
up
er
v
is
ed
L
ea
r
nin
g
T
h
e
s
e
m
i
-
s
u
p
er
v
i
s
ed
lear
n
i
n
g
p
ar
ad
ig
m
i
s
o
f
s
p
ec
ial
s
ig
n
i
f
ican
ce
i
n
b
o
th
t
h
eo
r
y
a
n
d
ap
p
licatio
n
s
.
I
n
m
a
n
y
M
L
ap
p
licatio
n
s
i
n
c
lu
d
in
g
ASR
,
u
n
lab
elled
d
ata
is
ab
u
n
d
an
t
b
u
t
lab
elli
n
g
is
e
x
p
en
s
iv
e
a
n
d
ti
m
e
-
co
n
s
u
m
i
n
g
.
I
t
is
p
o
s
s
ib
le
a
n
d
o
f
ten
h
elp
f
u
l
to
lev
er
a
g
e
in
f
o
r
m
atio
n
f
r
o
m
u
n
lab
el
led
d
ata
to
in
f
l
u
en
ce
lear
n
in
g
.
Se
m
i
-
s
u
p
er
v
is
ed
le
ar
n
in
g
i
s
tar
g
eted
at
p
r
ec
is
el
y
t
h
i
s
t
y
p
e
o
f
s
ce
n
ar
io
,
an
d
it
a
s
s
u
m
es
th
e
av
ailab
ilit
y
o
f
b
o
th
lab
elled
an
d
u
n
lab
elled
d
ata,
i.e
.
g
iv
e
n
b
y
eq
u
atio
n
6
:
(
6)
T
h
e
g
o
al
is
to
lev
er
ag
e
b
o
th
d
ata
s
o
u
r
ce
s
to
i
m
p
r
o
v
e
lear
n
i
n
g
p
er
f
o
r
m
an
ce
.
T
h
er
e
h
av
e
b
e
en
a
lar
g
e
n
u
m
b
er
o
f
s
e
m
i
-
s
u
p
er
v
is
ed
lear
n
in
g
alg
o
r
it
h
m
s
p
r
o
p
o
s
ed
in
th
e
liter
atu
r
e
an
d
v
ar
io
u
s
w
a
y
s
o
f
g
r
o
u
p
in
g
th
e
s
e
ap
p
r
o
ac
h
es.
Her
e
w
e
ca
teg
o
r
i
ze
s
e
m
i
-
s
u
p
er
v
i
s
ed
lear
n
in
g
m
et
h
o
d
s
b
ased
o
n
th
eir
in
d
u
c
tiv
e
o
r
tr
an
s
d
u
cti
v
e
n
atu
r
e.
T
h
e
k
e
y
d
if
f
er
en
ce
b
et
w
ee
n
in
d
u
cti
v
e
an
d
tr
an
s
d
u
ct
iv
e
lear
n
i
n
g
is
t
h
e
o
u
tco
m
e
o
f
lear
n
i
n
g
p
r
o
ce
s
s
.
I
n
th
e
f
o
r
m
er
s
etti
n
g
,
th
e
g
o
al
is
to
f
in
d
a
d
ec
is
io
n
f
u
n
cti
o
n
th
at
n
o
t
o
n
l
y
co
r
r
ec
tl
y
cla
s
s
i
f
ies
tr
ain
i
n
g
s
et
s
a
m
p
les,
b
u
t
also
g
e
n
er
a
lize
s
to
an
y
f
u
tu
r
e
s
a
m
p
le.
I
n
co
n
tr
ast,
tr
an
s
d
u
ctiv
e
lear
n
in
g
ai
m
s
at
d
ir
ec
tl
y
p
r
ed
ictin
g
t
h
e
o
u
tp
u
t
lab
el
s
o
f
a
test
s
et,
w
it
h
o
u
t
t
h
e
n
ee
d
o
f
g
en
er
alizi
n
g
to
o
th
er
s
a
m
p
le
s
.
I
n
th
is
r
e
g
ar
d
,
th
e
d
ir
ec
t
o
u
tco
m
e
o
f
tr
a
n
s
d
u
cti
v
e
s
e
m
i
-
s
u
p
er
v
is
ed
lear
n
i
n
g
is
a
s
et
o
f
lab
el
s
i
n
s
tead
o
f
a
d
ec
is
io
n
f
u
n
ctio
n
.
A
ll lea
r
n
i
n
g
p
ar
ad
ig
m
s
w
e
h
a
v
e
p
r
esen
ted
ar
e
in
d
u
c
tiv
e
i
n
n
atu
r
e.
1
.
4
.
4
.
Act
iv
e
L
ea
rning
A
cti
v
e
lear
n
i
n
g
is
a
s
i
m
i
lar
s
ettin
g
to
s
e
m
i
-
s
u
p
er
v
i
s
ed
lear
n
in
g
i
n
th
at,
in
ad
d
itio
n
to
a
s
m
all
a
m
o
u
n
t
o
f
lab
elled
d
at
a,
th
er
e
is
a
lar
g
e
a
m
o
u
n
t o
f
u
n
lab
elled
d
ata
av
ailab
le;
i.e
.
,
g
i
v
en
b
y
eq
u
atio
n
7
:
(
7
)
T
h
e
g
o
al
o
f
ac
ti
v
e
lear
n
i
n
g
,
h
o
w
e
v
er
,
is
to
q
u
er
y
t
h
e
m
o
s
t
in
f
o
r
m
ati
v
e
s
e
t
o
f
i
n
p
u
ts
to
b
e
lab
elled
,
h
o
p
in
g
to
i
m
p
r
o
v
e
cla
s
s
i
f
ica
tio
n
p
er
f
o
r
m
a
n
ce
w
it
h
t
h
e
m
i
n
i
m
u
m
n
u
m
b
er
o
f
q
u
er
ies
.
T
h
at
is
,
in
ac
ti
v
e
lear
n
in
g
,
t
h
e
lear
n
er
m
a
y
p
la
y
an
ac
ti
v
e
r
o
le
i
n
d
ec
id
i
n
g
t
h
e
d
ata
s
et
r
at
h
er
t
h
an
it b
e
p
as
s
i
v
el
y
g
iv
e
n
.
T
h
e
k
e
y
th
o
u
g
h
t
b
eh
in
d
ac
tiv
e
lear
n
in
g
i
s
t
h
at
a
M
L
ca
lcu
la
tio
n
c
an
ac
co
m
p
l
is
h
m
o
r
e
n
o
te
w
o
r
th
y
e
x
ec
u
t
io
n
,
e.
g
.
,
h
ig
h
er
cla
s
s
i
f
icatio
n
ac
cu
r
ac
y
,
w
i
th
f
e
w
er
tr
ai
n
i
n
g
lab
els
i
f
i
t
is
allo
w
ed
to
ch
o
o
s
e
t
h
e
s
u
b
s
et
o
f
d
ata
t
h
at
h
as
lab
els.
A
n
ac
ti
v
e
lear
n
er
m
i
g
h
t
p
o
s
t
u
r
e
q
u
es
tio
n
s
o
r
d
in
ar
il
y
as
u
n
lab
elled
i
n
f
o
r
m
atio
n
ca
s
es
to
b
e
n
a
m
ed
r
eg
u
lar
l
y
b
y
a
h
u
m
an
.
Fo
r
th
i
s
r
ea
s
o
n
,
it
is
s
o
m
eti
m
e
s
ca
ll
ed
q
u
er
y
lear
n
i
n
g
.
A
cti
v
e
lea
r
n
in
g
i
s
all
ar
o
u
n
d
in
s
p
ir
ed
in
n
u
m
er
o
u
s
p
r
ese
n
t
d
ay
ML
is
s
u
e
s
w
h
er
e
u
n
lab
el
led
in
f
o
r
m
atio
n
m
ig
h
t
b
e
co
p
io
u
s
o
r
ef
f
o
r
tles
s
l
y
g
o
tten
y
et
n
a
m
e
s
ar
e
tr
o
u
b
leso
m
e
ted
io
u
s
,
o
r
ex
p
en
s
i
v
e
to
o
b
tain
.
T
h
is
is
t
h
e
s
it
u
atio
n
f
o
r
s
p
ee
ch
r
ec
o
g
n
itio
n
.
B
r
o
ad
ly
,
ac
ti
v
e
lear
n
in
g
co
m
e
s
i
n
t
w
o
f
o
r
m
s
:
b
atc
h
ac
ti
v
e
le
a
r
n
in
g
,
w
h
er
e
a
s
u
b
s
e
t
o
f
d
ata
is
c
h
o
s
e
n
,
a
p
r
io
r
i
in
a
b
atch
to
b
e
lab
elled
.
T
h
e
lab
els
o
f
th
e
i
n
s
ta
n
ce
s
i
n
t
h
e
b
atch
ch
o
s
e
n
to
b
e
lab
elled
m
a
y
n
o
t,
u
n
d
er
th
is
ap
p
r
o
ac
h
,
in
f
l
u
en
ce
o
t
h
er
i
n
s
tan
ce
s
to
b
e
s
elec
ted
s
i
n
ce
al
l
in
s
tan
ce
s
ar
e
ch
o
s
e
n
at
o
n
c
e.
I
n
o
n
li
n
e
ac
ti
v
e
lear
n
in
g
,
o
n
th
e
o
t
h
er
h
an
d
,
i
n
s
ta
n
ce
s
ar
e
ch
o
s
e
n
o
n
e
-
by
-
o
n
e,
an
d
th
e
tr
u
e
lab
els
o
f
all
p
r
ev
io
u
s
l
y
lab
elled
in
s
ta
n
ce
s
m
a
y
b
e
u
s
ed
to
s
el
ec
t
o
th
er
in
s
tan
ce
s
to
b
e
lab
elled
.
Fo
r
th
is
r
ea
s
o
n
,
o
n
li
n
e
ac
tiv
e
lear
n
in
g
i
s
s
o
m
eti
m
es c
o
n
s
id
er
ed
m
o
r
e
p
o
w
er
f
u
l.
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
:
66
–
73
70
1
.
5
.
Art
if
icia
l N
eura
l N
et
w
o
rk
An
ar
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
is
a
n
i
n
ter
co
n
n
ec
ted
g
r
o
u
p
o
f
n
o
d
es,
d
is
ta
n
tl
y
r
elate
d
to
th
e
v
as
t
n
et
w
o
r
k
o
f
n
eu
r
o
n
s
i
n
a
b
r
ain
s
h
o
w
n
i
n
F
i
g
u
r
e
1
.
Her
e,
ea
ch
cir
cu
lar
n
o
d
e
r
ep
r
esen
ts
a
n
ar
tif
icia
l
n
e
u
r
o
n
a
n
d
an
ar
r
o
w
r
ep
r
esen
t
s
a
co
n
n
ec
t
io
n
f
r
o
m
t
h
e
o
u
tp
u
t
o
f
o
n
e
n
e
u
r
o
n
to
t
h
e
in
p
u
t
o
f
an
o
t
h
er
s
h
o
u
ld
(
id
ea
ll
y
)
b
e
ab
le
to
h
a
n
d
le
th
is
.
A
r
ti
f
icia
l
n
eu
r
al
n
et
w
o
r
k
co
n
s
is
t
o
f
th
r
e
e
t
y
p
e
o
f
la
y
er
n
a
m
el
y
i
n
p
u
t
l
a
y
er
,
h
id
d
en
la
y
er
an
d
o
u
tp
u
t la
y
er
.
Hid
d
en
la
y
er
is
co
n
n
ec
ted
b
et
w
ee
n
in
p
u
t a
n
d
o
u
tp
u
t la
y
er
.
Fig
u
r
e
1
.
A
r
ti
f
icial
Ne
u
r
al
Net
w
o
r
k
A
r
c
h
itect
u
r
e
1
.
6
.
Co
nv
o
lutio
na
l N
eura
l N
et
w
o
rks
C
NNs
ar
e
a
f
a
m
il
y
o
f
m
u
lti
-
l
a
y
er
n
eu
r
al
n
e
t
w
o
r
k
s
s
h
o
w
n
i
n
F
ig
u
r
e
2
p
ar
ticu
lar
l
y
d
esig
n
ed
f
o
r
u
s
e
o
n
t
w
o
-
d
i
m
en
s
io
n
al
d
ata,
s
u
c
h
as
i
m
ag
e
s
an
d
v
id
eo
s
.
C
N
Ns
ar
e
in
f
lu
e
n
ce
d
b
y
ea
r
lier
w
o
r
k
i
n
ti
m
e
-
d
ela
y
n
eu
r
al
n
et
w
o
r
k
s
(
T
DNN)
,
w
h
i
ch
r
ed
u
ce
lear
n
i
n
g
co
m
p
u
tatio
n
r
eq
u
ir
e
m
e
n
ts
b
y
s
h
ar
i
n
g
w
ei
g
h
t
s
i
n
a
te
m
p
o
r
al
d
i
m
en
s
io
n
a
n
d
ar
e
in
ten
d
ed
f
o
r
s
p
ee
ch
an
d
ti
m
e
-
s
er
ies
p
r
o
ce
s
s
i
n
g
.
C
NNs
ar
e
th
e
f
ir
s
t
tr
u
l
y
s
u
c
ce
s
s
f
u
l
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
w
h
er
e
m
an
y
la
y
er
s
o
f
a
h
ier
ar
c
h
y
w
er
e
s
u
cc
es
s
f
u
ll
y
tr
ain
ed
in
a
r
o
b
u
s
t
m
a
n
n
er
.
A
C
NN
is
a
ch
o
ice
o
f
ar
c
h
itect
u
r
e
t
h
at
l
ev
er
ag
es
s
p
atial
a
n
d
te
m
p
o
r
al
r
elatio
n
s
h
ip
s
to
r
ed
u
ce
t
h
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
w
h
ic
h
m
u
s
t
b
e
lear
n
ed
an
d
t
h
u
s
i
m
p
r
o
v
es
u
p
o
n
g
en
er
al
f
ee
d
-
f
o
r
w
ar
d
b
ac
k
p
r
o
p
ag
atio
n
tr
ain
i
n
g
.
C
NN
s
w
er
e
p
r
o
p
o
s
ed
as
a
d
ee
p
lear
n
in
g
f
r
a
m
e
w
o
r
k
th
a
t
is
m
o
ti
v
ate
d
b
y
m
i
n
i
m
al
d
a
ta
p
r
e
-
p
r
o
ce
s
s
i
n
g
r
eq
u
ir
e
m
e
n
ts
.
I
n
C
NN
s
,
s
m
all
p
o
r
tio
n
s
o
f
t
h
e
i
m
ag
e
ar
e
tr
ea
ted
as
in
p
u
t
s
to
th
e
lo
w
e
s
t
la
y
er
o
f
t
h
e
h
i
er
ar
ch
ical
s
tr
u
ctu
r
e.
I
n
f
o
r
m
a
tio
n
g
e
n
er
all
y
p
r
o
p
ag
ates
t
h
r
o
u
g
h
t
h
e
d
i
f
f
er
e
n
t
la
y
er
s
o
f
t
h
e
n
et
w
o
r
k
w
h
er
eb
y
a
t
ea
ch
la
y
er
d
ig
ital
f
ilter
i
n
g
i
s
ap
p
lied
i
n
o
r
d
er
to
o
b
tain
s
alie
n
t
f
ea
t
u
r
es
o
f
th
e
d
ata
o
b
s
er
v
ed
.
T
h
e
m
et
h
o
d
p
r
o
v
id
es
a
le
v
el
o
f
in
v
ar
ia
n
ce
to
s
h
i
f
t,
s
ca
le
a
n
d
r
o
tatio
n
as
t
h
e
lo
ca
l
r
ec
ep
tiv
e
f
ield
allo
w
s
th
e
n
e
u
r
o
n
o
r
p
r
o
c
ess
i
n
g
u
n
it
ac
ce
s
s
to
ele
m
en
tar
y
f
ea
tu
r
es
s
u
c
h
as
o
r
ien
ted
ed
g
es o
r
co
r
n
er
s
.
Fig
u
r
e
2
.
C
o
n
v
o
l
u
tio
n
al
Neu
r
a
l N
et
w
o
r
k
(
C
NN)
A
r
ch
i
tectu
r
e
[
1
2
]
1
.
7
.
Dee
p B
elief
Net
w
o
rk
s
DB
Ns
ar
e
co
m
p
o
s
ed
o
f
s
e
v
er
al
la
y
er
s
o
f
R
e
s
tr
icted
B
o
ltz
m
an
n
Ma
c
h
i
n
es,
a
t
y
p
e
o
f
n
e
u
r
al
n
et
w
o
r
k
s
h
o
w
n
i
n
F
ig
u
r
e
3
.
T
h
ese
n
et
w
o
r
k
s
ar
e
“
r
es
tr
icted
”
to
a
s
i
n
g
le
v
is
ib
le
la
y
er
a
n
d
s
in
g
le
h
id
d
en
la
y
er
,
w
h
er
e
co
n
n
ec
tio
n
s
ar
e
f
o
r
m
ed
b
et
w
ee
n
th
e
la
y
er
s
(
u
n
its
w
it
h
i
n
a
lay
er
ar
e
n
o
t
co
n
n
ec
ted
)
.
T
h
e
h
id
d
en
u
n
its
ar
e
tr
ain
ed
to
ca
p
tu
r
e
h
i
g
h
er
-
o
r
d
er
d
ata
co
r
r
elatio
n
s
t
h
at
ar
e
o
b
s
er
v
ed
at
t
h
e
v
is
ib
le
u
n
it
s
.
I
n
i
ti
all
y
,
a
s
id
e
f
r
o
m
t
h
e
to
p
t
w
o
la
y
er
s
,
w
h
ich
f
o
r
m
an
ass
o
ciati
v
e
m
e
m
o
r
y
,
d
ir
ec
ted
to
p
-
d
o
w
n
g
e
n
er
ati
v
e
w
ei
g
h
t
ar
e
u
s
ed
to
co
n
n
ec
t
th
e
la
y
er
s
o
f
a
DB
N.
R
B
Ms
ar
e
attr
ac
tiv
e
as
a
b
u
ild
in
g
b
lo
ck
,
o
v
er
m
o
r
e
tr
ad
itio
n
al
an
d
d
ee
p
ly
la
y
er
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
Dee
p
Ma
ch
in
e
lea
r
n
in
g
a
n
d
N
eu
r
a
l Netw
o
r
k
s
:
A
n
Ove
r
view
(
C
h
a
n
d
r
a
h
a
s
Mis
h
r
a
)
71
s
ig
m
o
id
b
elief
n
et
w
o
r
k
s
,
d
u
e
t
o
th
eir
ea
s
e
o
f
le
ar
n
in
g
t
h
ese
c
o
n
n
ec
tio
n
w
ei
g
h
t
s
,
th
e
in
i
tial
p
r
e
-
tr
ain
i
n
g
o
cc
u
r
s
in
a
n
u
n
s
u
p
er
v
is
ed
g
r
e
ed
y
la
y
er
-
by
-
la
y
er
m
a
n
n
er
to
o
b
tain
g
en
er
ati
v
e
w
eig
h
t
s
,
en
ab
led
b
y
w
h
at
Hin
to
n
h
as
ter
m
ed
co
n
tr
ast
iv
e
d
i
v
er
g
e
n
c
e.
Du
r
in
g
t
h
i
s
tr
ai
n
in
g
p
h
ase,
a
v
ec
to
r
v
is
p
r
ese
n
ted
to
t
h
e
v
is
ib
le
u
n
it
s
t
h
at
f
o
r
w
ar
d
v
alu
e
s
to
t
h
e
h
id
d
en
u
n
its
.
Go
in
g
i
n
r
ev
er
s
e,
t
h
e
v
is
i
b
l
e
u
n
i
t i
n
p
u
ts
ar
e
th
e
n
s
to
ch
a
s
ticall
y
f
o
u
n
d
in
a
n
atte
m
p
t to
r
ec
o
n
s
tr
u
ct
t
h
e
o
r
ig
in
al
i
n
p
u
t.
Fi
n
all
y
,
t
h
ese
n
e
w
v
is
ib
le
n
e
u
r
o
n
ac
t
iv
at
io
n
s
ar
e
f
o
r
w
ar
d
ed
s
u
c
h
t
h
at
o
n
e
s
tep
r
ec
o
n
s
tr
u
ctio
n
h
id
d
en
u
n
it
ac
tiv
a
tio
n
s
,
h
,
ca
n
b
e
attain
ed
.
P
er
f
o
r
m
i
n
g
th
ese
b
ac
k
an
d
f
o
r
th
s
tep
s
is
a
p
r
o
ce
s
s
k
n
o
w
n
as
Gib
b
s
s
a
m
p
lin
g
,
an
d
t
h
e
d
if
f
er
e
n
ce
in
th
e
co
r
r
elatio
n
o
f
th
e
h
id
d
en
ac
t
i
v
atio
n
s
an
d
v
is
ib
le
in
p
u
t
s
f
o
r
m
s
t
h
e
b
asi
s
f
o
r
a
w
eig
h
t
u
p
d
ate.
T
r
ain
in
g
ti
m
e
i
s
s
ig
n
i
f
ica
n
tl
y
r
ed
u
ce
d
as
it
ca
n
b
e
s
h
o
w
n
th
at
o
n
l
y
a
s
in
g
le
s
tep
is
n
ee
d
ed
to
ap
p
r
o
x
i
m
ate
m
a
x
i
m
u
m
li
k
eli
h
o
o
d
lear
n
in
g
.
E
ac
h
la
y
er
ad
d
ed
to
th
e
n
et
w
o
r
k
ca
n
i
m
p
r
o
v
es
th
e
lo
g
p
r
o
b
ab
ilit
y
o
f
t
h
e
tr
ai
n
i
n
g
d
ata,
w
h
ic
h
w
e
c
an
b
e
t
h
in
k
o
f
as
i
n
cr
ea
s
i
n
g
tr
u
e
r
ep
r
esen
tat
io
n
al
p
o
w
er
o
f
n
et
w
o
r
k
.
T
h
is
m
ea
n
in
g
f
u
l
ex
p
an
s
io
n
,
i
n
co
n
j
u
n
ct
io
n
w
ith
t
h
e
u
ti
lizatio
n
o
f
u
n
lab
elled
d
ata,
is
a
cr
itical
co
m
p
o
n
en
t i
n
a
n
y
d
ee
p
lear
n
in
g
ap
p
licatio
n
.
A
t
t
h
e
to
p
t
w
o
la
y
er
s
,
t
h
e
w
ei
g
h
t
s
ar
e
tied
to
g
e
th
er
,
s
u
c
h
th
at
th
e
o
u
tp
u
t
o
f
t
h
e
lo
w
er
la
y
e
r
s
p
r
o
v
id
e
s
a
r
ef
er
en
ce
cl
u
e
o
r
lin
k
f
o
r
th
e
to
p
la
y
er
to
“
as
s
o
ciate
”
w
it
h
i
ts
m
e
m
o
r
y
co
n
te
n
t
s
.
W
e
o
f
ten
e
n
co
u
n
ter
p
r
o
b
lem
s
w
h
er
e
d
is
cr
i
m
in
at
iv
e
p
er
f
o
r
m
a
n
ce
is
o
f
u
lt
i
m
a
te
co
n
ce
r
n
,
e.
g
.
in
cla
s
s
i
f
icat
io
n
task
s
.
A
DB
N
m
a
y
b
e
f
in
e
-
t
u
n
ed
af
ter
p
r
e
-
tr
ain
i
n
g
f
o
r
i
m
p
r
o
v
ed
d
is
cr
i
m
in
a
tiv
e
p
er
f
o
r
m
a
n
ce
b
y
u
tili
zi
n
g
lab
elled
d
ata
th
r
o
u
g
h
b
ac
k
-
p
r
o
p
ag
atio
n
.
A
t
t
h
is
p
o
in
t,
a
s
et
o
f
lab
el
s
is
attac
h
ed
to
th
e
to
p
lay
er
(
ex
p
an
d
i
n
g
t
h
e
as
s
o
ciativ
e
m
e
m
o
r
y
)
to
clar
if
y
ca
teg
o
r
y
b
o
u
n
d
ar
ies
in
t
h
e
n
et
w
o
r
k
t
h
r
o
u
g
h
w
h
ich
a
n
e
w
s
et
o
f
b
o
tto
m
-
u
p
,
r
ec
o
g
n
it
io
n
w
ei
g
h
ts
ar
e
lear
n
ed
.
I
t
h
as
b
ee
n
s
h
o
w
n
in
th
at
s
u
c
h
n
e
t
w
o
r
k
s
o
f
te
n
p
er
f
o
r
m
b
etter
th
an
t
h
o
s
e
tr
ai
n
ed
ex
clu
s
i
v
el
y
w
it
h
b
ac
k
p
r
o
p
ag
atio
n
.
T
h
is
m
a
y
b
e
i
n
t
u
iti
v
el
y
e
x
p
lain
ed
b
y
t
h
e
f
ac
t t
h
at
b
ac
k
-
p
r
o
p
ag
atio
n
Fo
r
DB
Ns
is
o
n
l
y
r
eq
u
ir
ed
to
p
er
f
o
r
m
a
lo
ca
l
s
ea
r
ch
o
n
th
e
w
ei
g
h
t
(
p
ar
am
e
ter
)
s
p
ac
e,
s
p
e
ed
in
g
tr
ain
i
n
g
a
n
d
co
n
v
er
g
en
ce
ti
m
e
in
r
elatio
n
to
tr
ad
itio
n
al
f
ee
d
-
f
o
r
w
ar
d
n
e
u
r
al
n
et
w
o
r
k
s
Fig
u
r
e
3
.
Dee
p
B
elief
Neu
r
al
Net
w
o
r
k
(
DB
NN)
A
r
c
h
itect
u
r
e
[
1
3
]
2.
I
M
P
L
E
M
E
NT
AT
I
O
N
T
E
C
H
NIQ
U
E
S
L
ea
r
n
i
n
g
ca
n
b
e
i
m
p
le
m
e
n
t
ed
b
y
v
ar
io
u
s
m
et
h
o
d
o
lo
g
y
o
r
tech
n
iq
u
e
s
d
ep
en
d
in
g
u
p
o
n
th
e
r
eq
u
ir
e
m
en
t.
B
asicall
y
lear
n
i
n
g
ca
n
b
e
ca
teg
o
r
ized
in
t
w
o
t
y
p
es:
1.
S
u
p
er
v
i
s
ed
lear
n
i
n
g
2.
U
n
s
u
p
er
v
i
s
ed
lear
n
i
n
g
I
n
s
u
p
er
v
is
ed
lear
n
i
n
g
t
w
o
tec
h
n
iq
u
e
is
u
s
ed
1.
R
eg
r
es
s
io
n
2.
C
las
s
i
f
icatio
n
W
h
ile
u
n
s
u
p
er
v
i
s
ed
lear
n
in
g
i
s
i
m
p
le
m
e
n
ted
b
y
clu
s
ter
in
g
al
g
o
r
ith
m
s
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
:
66
–
73
72
3.
AP
P
L
I
CA
T
I
O
N
S
A
cc
o
r
d
in
g
to
E
.
Do
n
B
o
x
et
al.
[
1
]
n
eu
r
al
n
et
w
o
r
k
s
m
o
d
el
ca
n
b
e
u
s
ed
f
o
r
d
em
a
n
d
f
o
r
ec
asti
n
g
i
n
d
er
eg
u
lated
e
n
v
ir
o
n
m
en
t.
Neu
r
al
n
et
w
o
r
k
s
ar
e
d
e
s
ig
n
ed
a
n
d
tr
ain
ed
o
n
t
h
e
b
as
is
o
f
a
g
g
r
eg
ate
d
e
m
a
n
d
s
o
f
t
h
e
g
r
o
u
p
s
o
f
s
u
r
v
e
y
ed
cu
s
to
m
er
s
o
f
d
if
f
er
en
t c
ate
g
o
r
ies.
T
h
e
m
o
s
t
f
r
eq
u
e
n
tl
y
e
n
co
u
n
ter
ed
d
ec
is
io
n
m
a
k
i
n
g
tas
k
s
o
f
h
u
m
a
n
ac
ti
v
it
y
i
s
cl
ass
i
f
icatio
n
.
C
las
s
i
f
icatio
n
p
r
o
b
le
m
o
cc
u
r
s
w
h
e
n
an
o
b
j
ec
t
n
ee
d
s
to
b
e
as
s
ig
n
ed
i
n
to
a
p
r
ed
ef
in
ed
cla
s
s
b
ased
o
n
a
n
u
m
b
er
o
f
o
b
s
er
v
ed
attr
ib
u
tes
r
elate
d
to
th
at
o
b
j
ec
t.
Ma
n
y
p
r
o
b
le
m
s
in
b
u
s
in
e
s
s
,
s
cie
n
ce
,
i
n
d
u
s
tr
y
,
an
d
m
ed
icin
e
ca
n
b
e
tr
ea
ted
as
cla
s
s
i
f
icat
io
n
p
r
o
b
lem
s
.
E
x
a
m
p
les
in
c
lu
d
e
b
an
k
r
u
p
tc
y
p
r
ed
ictio
n
,
cr
ed
it
s
co
r
in
g
,
m
ed
ica
l
d
iag
n
o
s
i
s
,
q
u
alit
y
co
n
tr
o
l,
h
a
n
d
w
r
itte
n
ch
ar
ac
ter
r
ec
o
g
n
itio
n
,
an
d
s
p
ee
ch
r
ec
o
g
n
it
io
n
.
[
2
]
Neu
r
al
n
et
w
o
r
k
an
d
g
e
n
etic
a
lg
o
r
ith
m
s
ar
e
u
s
ed
f
o
r
w
eb
m
i
n
in
g
.
Sa
n
k
ar
K.
P
al
et.
al
[
3
]
d
escr
ib
e
w
eb
m
in
i
n
g
in
s
o
f
t c
o
m
p
u
ti
n
g
f
r
a
m
e
w
o
r
k
.
So
f
t
co
m
p
u
ti
n
g
p
ar
ad
ig
m
li
k
e
f
u
zz
y
s
e
ts
(
FS
)
,
ar
ti
f
icial
n
eu
r
al
n
et
w
o
r
k
s
(
A
N
N)
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
s
(
SVMs)
i
s
u
s
ed
i
n
B
io
in
f
o
r
m
atics.
[
4
]
T
h
e
r
esear
ch
co
m
m
u
n
it
y
h
a
s
s
tatr
ted
lo
o
k
in
g
f
o
r
I
P
tr
af
f
ic
class
i
f
icatio
n
tech
n
iq
u
es
t
h
at
d
o
n
o
t
r
ely
o
n
„
w
e
ll
k
n
o
w
n
‟
T
C
P
o
r
UDP
p
o
r
t
n
u
m
b
er
s
,
o
r
in
ter
p
r
eti
n
g
t
h
e
co
n
te
n
t
s
o
f
p
ac
k
et
p
a
y
l
o
ad
s
.
Ne
w
w
o
r
k
is
e
m
er
g
i
n
g
o
n
th
e
u
s
e
o
f
s
tati
s
tical
tr
af
f
ic
c
h
ar
ac
ter
is
tics
t
o
ass
is
t
i
n
th
e
id
en
t
if
icatio
n
an
d
class
i
f
icatio
n
p
r
o
ce
s
s
.
T
h
is
s
u
r
v
e
y
p
ap
er
lo
o
k
s
at
e
m
er
g
in
g
r
esear
c
h
i
n
to
th
e
ap
p
licatio
n
o
f
Ma
ch
i
n
e
L
ea
r
n
i
n
g
(
M
L
)
tech
n
iq
u
es
to
I
P
tr
af
f
ic
clas
s
if
ica
tio
n
-
an
in
ter
-
d
is
cip
li
n
ar
y
b
le
n
d
o
f
I
P
n
et
w
o
r
k
i
n
g
an
d
d
ata
m
i
n
i
n
g
tech
n
iq
u
es [
5
]
I
t
is
cr
u
cial
f
o
r
f
i
n
a
n
cial
in
s
tit
u
tio
n
s
,
th
e
ab
ilit
y
to
p
r
ed
ict
o
r
f
o
r
ec
ast
b
u
s
i
n
ess
f
ail
u
r
es,
as
in
co
r
r
ec
t
d
ec
is
io
n
s
ca
n
h
a
v
e
d
ir
ec
t
f
i
n
an
cia
l
co
n
s
eq
u
e
n
ce
s
.
T
h
er
e
ar
e
th
e
t
w
o
m
aj
o
r
r
e
s
ea
r
ch
p
r
o
b
le
m
s
i
n
t
h
e
ac
co
u
n
ti
n
g
an
d
f
i
n
a
n
ce
d
o
m
a
in
ar
e
B
an
k
r
u
p
tc
y
p
r
ed
ictio
n
an
d
cr
ed
it
s
co
r
in
g
.
I
n
t
h
e
l
iter
atu
r
e,
a
n
u
m
b
er
o
f
m
o
d
el
s
h
a
v
e
b
ee
n
d
ev
elo
p
ed
to
p
r
e
d
ict
w
h
e
th
er
b
o
r
r
o
w
er
s
ar
e
in
d
an
g
er
o
f
b
an
k
r
u
p
tc
y
a
n
d
w
h
e
th
er
t
h
e
y
s
h
o
u
ld
b
e
co
n
s
id
er
ed
a
g
o
o
d
o
r
b
ad
cr
ed
it
r
is
k
.
Si
n
ce
t
h
e
1
9
9
0
s
,
m
ac
h
i
n
e
-
lear
n
i
n
g
tec
h
n
i
q
u
es,
s
u
c
h
as
n
e
u
r
al
n
et
w
o
r
k
s
an
d
d
ec
is
io
n
tr
ee
s
,
h
av
e
b
ee
n
s
t
u
d
ied
ex
te
n
s
i
v
el
y
a
s
to
o
ls
f
o
r
b
an
k
r
u
p
tc
y
p
r
ed
icti
o
n
an
d
cr
ed
it sco
r
e
m
o
d
ell
in
g
.
[
6
]
L
ea
r
n
i
n
g
m
et
h
o
d
s
th
at
h
av
e
b
ee
n
ap
p
lied
to
C
R
s
class
i
f
y
in
g
th
e
m
u
n
d
er
s
u
p
er
v
is
ed
an
d
u
n
s
u
p
er
v
is
ed
lear
n
in
g
.
So
m
e
o
f
t
h
e
m
o
s
t
i
m
p
o
r
tan
t,
a
n
d
co
m
m
o
n
l
y
u
s
ed
,
lear
n
in
g
al
g
o
r
i
th
m
s
w
as
p
r
o
v
id
ed
alo
n
g
w
it
h
t
h
eir
ad
v
a
n
tag
e
s
an
d
d
is
ad
v
an
ta
g
es a
r
e
d
is
cu
s
s
ed
in
th
is
liter
at
u
r
e.
[
7
]
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
Dee
p
Ma
ch
in
e
lea
r
n
in
g
a
n
d
N
eu
r
a
l Netw
o
r
k
s
:
A
n
Ove
r
view
(
C
h
a
n
d
r
a
h
a
s
Mis
h
r
a
)
73
4.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
a
d
e
ep
d
is
cu
s
s
io
n
ab
o
u
t
m
ac
h
in
e
lear
n
in
g
m
et
h
o
d
s
an
d
th
eir
i
m
p
le
m
en
tatio
n
h
as
b
ee
n
d
is
cu
s
s
ed
.
I
t
is
clea
r
l
y
s
h
o
w
n
th
at
d
if
f
er
en
t
m
et
h
o
d
s
u
s
es
d
if
f
er
e
n
t
al
g
o
r
ith
m
f
o
r
i
m
p
le
m
en
tatio
n
.
I
t
is
also
co
n
clu
d
ed
th
a
t
Ne
u
r
al
Net
w
o
r
k
an
d
S
u
p
p
o
r
t
v
ec
to
r
m
ac
h
i
n
e
is
m
o
s
t
p
o
p
u
lar
tech
n
iq
u
e
s
to
i
m
p
le
m
e
n
t
t
h
e
m
ac
h
in
e
lear
n
i
n
g
p
ar
ad
ig
m
.
D
ee
p
lear
n
in
g
is
e
x
ten
d
ed
v
er
s
i
o
n
o
f
s
u
p
er
v
is
ed
lear
n
in
g
.
I
t
is
f
i
n
all
y
co
n
cl
u
d
ed
th
at
C
o
n
v
o
lu
t
io
n
n
eu
r
al
n
et
w
o
r
k
an
d
Dee
p
B
elie
f
n
et
w
o
r
k
ar
e
t
w
o
p
o
w
er
f
u
l te
c
h
n
iq
u
e
s
w
h
ich
m
a
y
b
e
u
s
ed
to
s
o
lv
e
v
a
r
io
u
s
co
m
p
lex
p
r
o
b
lem
s
u
s
i
n
g
d
ee
p
lear
n
in
g
.
Dee
p
lear
n
in
g
p
lat
f
o
r
m
s
ca
n
also
b
e
b
en
ef
ited
f
r
o
m
en
g
i
n
ee
r
ed
f
ea
t
u
r
es
w
h
ile
lea
r
n
in
g
m
o
r
e
co
m
p
le
x
r
ep
r
esen
tatio
n
s
w
h
ic
h
en
g
i
n
ee
r
ed
s
y
s
t
e
m
s
t
y
p
icall
y
lack
.
I
t is ab
u
n
d
a
n
tl
y
clea
r
t
h
at
ad
v
an
ce
m
e
n
t
s
m
ad
e
w
it
h
r
e
s
p
e
ct
to
d
ev
elo
p
in
g
d
ee
p
m
ac
h
i
n
e
le
ar
n
in
g
s
y
s
te
m
s
w
ill
u
n
d
o
u
b
ted
l
y
s
h
ap
e
th
e
f
u
t
u
r
e
o
f
m
ac
h
i
n
e
lear
n
i
n
g
a
n
d
ar
tif
i
cial
in
tell
ig
e
n
ce
s
y
s
te
m
s
in
g
e
n
er
al.
RE
F
E
R
E
NC
E
S
[1
]
Ch
a
ry
to
n
iu
k
W
ik
to
r
,
Bo
x
E.
Do
n
,
L
e
e
W
e
i
-
J
e
n
,
M
o
-
S
h
in
g
,
Ko
ta
s
Ch
e
n
P
a
u
l
a
n
d
Olin
d
a
P
e
ter
V
a
n
,
“
Ne
u
ra
l
-
Ne
tw
o
rk
-
B
a
se
d
De
m
a
n
d
F
o
re
c
a
stin
g
in
De
re
g
u
late
d
En
v
iro
n
m
e
n
t,
”
IEE
E
T
ra
n
s
.
On
In
d
u
stry
Ap
p
li
c
a
ti
o
n
s,
Vo
l.
3
6
,
No
.
3
,
M
a
y
/Ju
n
e
2
0
0
0
.
[2
]
Zh
a
n
g
G
u
o
q
ian
g
P
e
ter,
“
Ne
u
ra
l
Ne
tw
o
rk
s
f
o
r
Cla
ss
i
f
ic
a
ti
o
n
:
A
S
u
rv
e
y
,
”
IEE
E
T
ra
n
s.
On
S
y
st
e
ms
,
M
a
n
,
An
d
Cy
b
e
rn
a
ti
c
s
—
P
a
rt C:
A
p
p
li
c
a
ti
o
n
s a
n
d
re
v
iews
,
Vo
l.
3
0
,
No
.
4
,
N
o
v
e
mb
e
r
2
0
0
0
.
[3
]
S
a
n
k
a
r
K.
P
a
l
,
V
a
ru
n
T
a
lw
a
r
,
a
n
d
P
a
b
it
ra
M
it
ra
,
“
W
e
b
M
in
in
g
in
S
o
f
t
Co
m
p
u
ti
n
g
F
ra
m
e
w
o
r
k
:
Re
l
e
v
a
n
c
e
,
S
tate
o
f
th
e
A
rt
a
n
d
F
u
tu
re
Dire
c
ti
o
n
s,
”
I
EE
E
T
ra
n
s.On
Ne
u
ra
l
Ne
tw
o
rk
s,
Vo
l.
1
3
,
NO.
5
,
S
e
p
tem
b
e
r 2
0
0
2
.
[4
]
S
u
sh
m
it
a
M
it
ra
,
Yo
ich
i
Ha
y
a
sh
i
,
“
Bio
in
f
o
rm
a
ti
c
s
w
it
h
S
o
f
t
C
o
m
p
u
ti
n
g
,
”
IEE
E
T
ra
n
s.
On
S
y
s
tem
,
M
a
n
a
n
d
Cy
b
e
rn
a
ti
c
s
—
P
a
rt C:
A
p
p
li
c
a
ti
o
n
a
n
d
Rev
iews
,
Vo
l.
3
6
,
N
o
.
5
,
S
e
p
tem
b
e
r
2
0
0
6
.
[5
]
T
h
u
y
T
.
T
.
Ng
u
y
e
n
a
n
d
G
re
n
v
il
l
e
A
r
m
it
a
g
e
,
“
A
S
u
rv
e
y
o
f
Tec
h
n
iq
u
e
s
f
o
r
In
tern
e
t
T
ra
ff
i
c
Clas
sif
ica
ti
o
n
u
sin
g
M
a
c
h
in
e
L
e
a
rn
in
g
,
”
IEE
E
C
o
mm
u
n
ica
ti
o
n
s
S
u
rv
e
y
s
&
T
u
to
ria
ls,
Vo
l.
1
0
,
No
.
4
,
Fo
u
rth
Qu
a
rte
r
2
0
0
8
.
[6
]
W
e
i
-
Ya
n
g
L
in
,
Ya
-
Ha
n
Hu
C
h
ih
-
F
o
n
g
T
sa
i,
“
M
a
c
h
in
e
L
e
a
rn
in
g
i
n
F
i
n
a
n
c
ial
Crisis
P
re
d
icti
o
n
:
A
S
u
rv
e
y
,
”
IEE
E
T
ra
n
s.
O
n
S
y
ste
m,
M
a
n
a
n
d
Cy
b
e
rn
a
ti
c
s
—
P
a
rt C:
A
p
p
l
ica
ti
o
n
a
n
d
Rev
iews
,
Vo
l.
4
2
,
No
.
4
,
J
u
ly
2
0
1
2
.
[7
]
M
a
rio
Bk
a
ss
in
y
Ya
n
g
L
i,
S
u
d
h
a
rm
a
n
K.
Ja
y
a
w
e
e
ra
,
“
A
S
u
rv
e
y
o
n
M
a
c
h
in
e
-
L
e
a
rn
in
g
T
e
c
h
n
iq
u
e
s
in
Co
g
n
it
iv
e
Ra
d
io
s,
”
IEE
E
Co
mm
u
n
ica
ti
o
n
s
S
u
rv
e
y
s
&
T
u
to
ria
ls
,
Vo
l.
1
5
,
No
.
3
,
T
h
ir
d
Qu
a
rter
2
0
1
3
.
[8
]
L
i
De
n
g
,
X
iao
L
i,
“
M
a
c
h
in
e
L
e
a
rn
in
g
P
a
ra
d
ig
m
s
f
o
r
S
p
e
e
c
h
Re
c
o
g
n
it
io
n
:
A
n
Ov
e
r
v
ie
w
,
”
IEE
E
T
ra
n
s
On
Au
d
io
,
S
p
e
e
c
h
,
A
n
d
L
a
n
g
u
a
g
e
Pro
c
e
ss
in
g
,
V
o
l.
2
1
,
NO.
5
,
M
a
y
2
0
1
3
.
[9
]
Da
izh
a
n
Ch
e
n
g
,
H
o
n
g
sh
e
n
g
Qi,
“
S
tate
–
S
p
a
c
e
A
n
a
l
y
sis
o
f
Bo
o
lea
n
Ne
tw
o
rk
s,”
IEE
E
T
ra
n
s
.
On
Ne
u
ra
l
Ne
tw
o
rk
s
,
Vo
l.
2
1
,
No
.
4
,
Ap
ri
l
2
0
1
0
.
[1
0
]
Yo
sh
u
a
Be
n
g
io
,
A
a
ro
n
Co
u
rv
il
le
a
n
d
P
a
sc
a
l
V
in
c
e
n
t,
“
Re
p
re
se
n
tati
o
n
L
e
a
rn
in
g
:
A
R
e
v
ie
w an
d
Ne
w
P
e
rsp
e
c
ti
v
e
s,
”
IEE
E
T
ra
n
s.
O
n
Pa
tt
e
rn
A
n
a
lys
is
and
M
a
c
h
i
n
e
In
telli
g
e
n
c
e
,
V
o
l
.
3
5
,
NO.
8
,
Au
g
u
st
2
0
1
3
.
[1
1
]
X
u
e
-
W
e
n
Ch
e
n
,
X
iao
t
o
n
g
L
in
,
“
Big
Da
ta
De
e
p
L
e
a
rn
in
g
:
Ch
a
ll
e
n
g
e
s
a
n
d
P
e
rsp
e
c
ti
v
e
s,”
Dig
it
a
l
Ob
jec
t
Id
e
n
ti
fi
e
r
1
0
.
1
1
0
9
/I
EE
E
ACC
ES
S
.
2
0
1
4
.
2
3
2
5
0
2
9
.
[1
2
]
h
tt
p
:
//
c
s2
3
1
n
.
g
it
h
u
b
.
i
o
/co
n
v
o
lu
t
io
n
a
l
-
n
e
tw
o
rk
s/#
o
v
e
rv
ie
w
[1
3
]
h
tt
p
:
//
im
a
g
e
.
slid
e
sh
a
re
c
d
n
.
c
o
m
/d
e
e
p
-
b
e
li
e
f
-
n
e
ts1
1
6
6
/9
5
/d
e
e
p
-
b
e
li
e
f
-
n
e
ts
-
3
-
7
2
8
.
j
p
g
?
c
b
=
1
2
7
2
2
8
2
8
2
5
APPENDI
X
Def
i
n
itio
n
s
o
f
a
S
u
b
s
et
o
f
C
o
m
m
o
n
l
y
Used
S
y
m
b
o
l
s
an
d
N
o
tatio
n
s
i
n
T
h
is
A
r
ticle
Evaluation Warning : The document was created with Spire.PDF for Python.