I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
25
,
p
p
.
1
9
4
9
~
1
9
6
0
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
2
.
pp
1
9
4
9
-
1
9
6
0
1949
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Rev
isiting
self
-
su
perv
ised co
ntras
ti
v
e learning
f
o
r im
ba
la
nced
cla
ss
ificatio
n
Xia
o
lin
g
G
a
o
1
,
2
,
M
uh
a
m
m
a
d
I
zz
a
d Ra
m
li
1
,
M
a
rsh
im
a
M
o
hd
Ro
s
li
1
,
3
,
Nursu
r
ia
t
i J
a
m
i
l
1
,
Sy
ed
M
o
hd
Z
a
hid
Sy
ed
Z
a
in
a
l A
rif
f
in
1
1
C
o
l
l
e
g
e
o
f
C
o
m
p
u
t
i
n
g
,
I
n
f
o
r
mat
i
c
s
a
n
d
M
a
t
h
e
m
a
t
i
c
s
,
U
n
i
v
e
r
si
t
i
Te
k
n
o
l
o
g
i
M
A
R
A
,
S
h
a
h
A
l
a
m
,
M
a
l
a
y
s
i
a
2
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
X
i
n
h
u
a
C
o
l
l
e
g
e
o
f
N
i
n
g
x
i
a
U
n
i
v
e
r
si
t
y
,
Y
i
n
c
h
u
a
n
,
C
h
i
n
a
3
I
n
st
i
t
u
t
e
f
o
r
P
a
t
h
o
l
o
g
y
,
L
a
b
o
r
a
t
o
r
y
a
n
d
F
o
r
e
n
s
i
c
M
e
d
i
c
i
n
e
(
I
-
P
P
e
r
F
o
r
M
)
,
U
n
i
v
e
r
si
t
i
Te
k
n
o
l
o
g
i
M
A
R
A
,
S
h
a
h
A
l
a
m
,
M
a
l
a
y
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
2
2
,
2
0
2
4
R
ev
is
ed
Oct
9
,
2
0
2
4
Acc
ep
ted
Oct
2
3
,
2
0
2
4
Clas
s
imb
a
lan
c
e
re
m
a
in
s
a
fo
rm
id
a
b
le
c
h
a
ll
e
n
g
e
in
m
a
c
h
in
e
lea
rn
in
g
,
p
a
rti
c
u
larly
a
ffe
c
ti
n
g
field
s
th
a
t
d
e
p
e
n
d
o
n
a
c
c
u
ra
te
c
las
sifica
ti
o
n
a
c
ro
ss
sk
e
we
d
d
a
tas
e
ts,
su
c
h
a
s
m
e
d
ica
l
ima
g
in
g
a
n
d
so
ftwa
re
d
e
fe
c
t
p
re
d
ictio
n
.
Trad
it
io
n
a
l
a
p
p
ro
a
c
h
e
s
o
fte
n
fa
il
to
a
d
e
q
u
a
tely
a
d
d
r
e
ss
th
e
u
n
d
e
rre
p
re
se
n
tati
o
n
o
f
m
i
n
o
ri
ty
c
las
se
s,
lea
d
in
g
to
m
o
d
e
ls
t
h
a
t
e
x
h
i
b
it
h
ig
h
p
e
rfo
rm
a
n
c
e
o
n
m
a
jo
rit
y
c
las
se
s
b
u
t
h
a
v
e
p
o
o
r
p
e
rf
o
rm
a
n
c
e
o
n
c
rit
ica
l
m
in
o
rit
y
c
las
se
s.
S
e
lf
-
su
p
e
rv
is
e
d
c
o
n
tras
ti
v
e
lea
rn
i
n
g
h
a
s
b
e
c
o
m
e
a
n
e
x
trem
e
ly
e
n
c
o
u
ra
g
i
n
g
m
e
th
o
d
fo
r
th
is
issu
e
,
e
n
a
b
li
n
g
t
h
e
u
ti
l
iza
ti
o
n
o
f
u
n
lab
e
l
e
d
d
a
ta
t
o
g
e
n
e
ra
te
ro
b
u
st
a
n
d
g
e
n
e
ra
li
z
a
b
le
m
o
d
e
ls.
Th
is
p
a
p
e
r
re
v
iew
s
th
e
a
d
v
a
n
c
e
m
e
n
ts
in
se
lf
-
su
p
e
rv
ise
d
c
o
n
tras
ti
v
e
le
a
rn
in
g
f
o
r
imb
a
lan
c
e
d
c
las
sifica
ti
o
n
,
fo
c
u
si
n
g
o
n
m
e
th
o
d
o
l
o
g
ies
th
a
t
e
n
h
a
n
c
e
m
o
d
e
l
p
e
rfo
rm
a
n
c
e
th
r
o
u
g
h
i
n
n
o
v
a
ti
v
e
c
o
n
tras
ti
v
e
l
o
ss
f
u
n
c
ti
o
n
s
a
n
d
d
a
ta
a
u
g
m
e
n
tatio
n
stra
teg
ies
.
By
p
u
l
li
n
g
sim
il
a
r
in
sta
n
c
e
s
c
lo
se
r
a
n
d
p
u
s
h
in
g
d
issim
il
a
r
o
n
e
s
a
p
a
rt,
th
e
se
tec
h
n
iq
u
e
s
h
e
l
p
m
it
i
g
a
te
th
e
b
ias
e
s
i
n
h
e
re
n
t
in
imb
a
lan
c
e
d
d
a
tas
e
ts.
We
c
rit
i
c
a
ll
y
a
n
a
ly
z
e
th
e
e
ffe
c
ti
v
e
n
e
ss
o
f
th
e
se
m
e
th
o
d
s
in
d
i
v
e
rse
sc
e
n
a
rio
s
a
n
d
p
ro
p
o
se
fu
t
u
re
re
se
a
rc
h
d
irec
ti
o
n
s
a
ime
d
a
t
re
fin
in
g
t
h
e
se
a
p
p
ro
a
c
h
e
s
f
o
r
b
ro
a
d
e
r
a
p
p
li
c
a
ti
o
n
in
re
a
l
-
wo
rld
se
tt
in
g
s.
Th
is
re
v
iew
se
rv
e
s
a
s
a
g
u
id
e
f
o
r
re
se
a
rc
h
e
rs
e
x
p
lo
r
in
g
t
h
e
p
o
ten
ti
a
l
o
f
c
o
n
tras
ti
v
e
lea
rn
i
n
g
t
o
a
d
d
re
ss
c
las
s
imb
a
lan
c
e
s,
h
ig
h
li
g
h
ti
n
g
re
c
e
n
t
su
c
c
e
ss
e
s a
n
d
id
e
n
ti
fy
in
g
c
ru
c
ial
g
a
p
s th
a
t
n
e
e
d
a
d
d
re
ss
in
g
.
K
ey
w
o
r
d
s
:
C
las
s
im
b
alan
ce
C
o
n
tr
asti
v
e
lear
n
in
g
I
m
b
alan
ce
d
class
if
icatio
n
Self
-
s
u
p
er
v
is
ed
lear
n
in
g
I
m
b
alan
ce
lear
n
in
g
R
ep
r
esen
tatio
n
lear
n
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mu
h
am
m
ad
I
zz
ad
R
am
li
C
o
lleg
e
o
f
C
o
m
p
u
tin
g
,
I
n
f
o
r
m
atics a
n
d
Ma
th
em
atics,
Un
iv
er
s
iti T
ek
n
o
lo
g
i M
AR
A
(
UiT
M)
Sh
ah
Alam
4
0
4
5
0
,
Selan
g
o
r
,
Ma
lay
s
ia
E
m
ail:
izza
d
r
am
li@
u
itm
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
C
las
s
im
b
alan
ce
s
ig
n
if
ican
tly
ch
allen
g
es
m
ac
h
in
e
lear
n
in
g
,
p
ar
ticu
lar
ly
in
f
ield
s
r
eq
u
i
r
in
g
p
r
ec
is
e
class
p
r
ed
ictio
n
s
s
u
ch
as
m
ed
ical
im
ag
in
g
,
im
ag
e
r
ec
o
g
n
iti
o
n
,
an
d
s
o
f
twar
e
d
ef
ec
t
p
r
e
d
i
ctio
n
[
1
]
.
I
n
th
ese
d
o
m
ain
s
,
th
e
d
is
p
r
o
p
o
r
tio
n
ate
r
ep
r
esen
tatio
n
o
f
class
es
o
f
t
en
b
iases
alg
o
r
ith
m
s
to
war
d
s
th
e
m
ajo
r
ity
class
,
co
m
p
r
o
m
is
in
g
th
e
d
etec
tio
n
o
f
less
p
r
ev
ale
n
t,
b
u
t
o
f
ten
m
o
r
e
cr
itical,
m
in
o
r
ity
class
es
[
2
]
.
T
r
ad
itio
n
al
class
if
ier
s
ty
p
ically
ex
ce
l
wit
h
b
alan
ce
d
d
atasets
b
u
t
s
tr
u
g
g
le
with
s
k
ewe
d
d
is
tr
ib
u
tio
n
s
,
lead
in
g
to
p
o
o
r
g
en
er
aliza
tio
n
o
f
r
ea
l
-
wo
r
ld
d
ata
[
3
]
,
[
4
]
.
T
o
cir
cu
m
v
e
n
t
th
e
lim
itatio
n
s
o
f
lab
elled
d
atasets
,
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
h
as
b
ec
o
m
e
a
r
eliab
le
o
p
tio
n
,
e
n
ab
lin
g
m
o
d
els
to
p
r
e
-
tr
ain
o
n
im
m
en
s
e
q
u
a
n
titi
es
o
f
u
n
lab
eled
d
ata
an
d
t
h
en
f
in
e
-
tu
n
e
t
h
em
o
n
s
m
aller
,
tag
g
ed
d
atasets
,
im
p
r
o
v
in
g
s
u
b
s
eq
u
en
t
task
p
e
r
f
o
r
m
an
ce
[
5
]
.
T
h
is
lear
n
i
n
g
p
ar
a
d
ig
m
is
ex
c
ep
tio
n
ally
v
alu
ab
le
wh
er
e
d
ata
an
n
o
tatio
n
is
co
s
tly
o
r
im
p
r
ac
tical
[
6
]
,
[
7
]
.
Am
o
n
g
v
ar
i
o
u
s
s
elf
-
s
u
p
er
v
is
ed
tech
n
iq
u
es,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
4
9
-
1
9
6
0
1950
co
n
tr
asti
v
e
lear
n
in
g
h
as
b
e
en
n
o
tab
ly
e
f
f
ec
tiv
e
in
m
a
n
ag
in
g
class
im
b
alan
ce
s
.
C
o
n
tr
asti
v
e
lear
n
in
g
tech
n
iq
u
es
h
elp
to
b
etter
r
ep
r
esen
t
u
n
d
e
r
-
r
ep
r
esen
ted
class
es
b
y
en
h
an
cin
g
th
e
p
r
o
ce
s
s
o
f
lear
n
in
g
r
ep
r
esen
tatio
n
s
b
y
p
u
llin
g
s
i
m
ilar
s
am
p
les
clo
s
er
an
d
p
u
s
h
in
g
d
is
s
im
ilar
o
n
es
ap
ar
t
in
th
e
f
ea
tu
r
e
s
p
ac
e,
wh
ich
h
as
s
h
o
wn
p
r
o
m
is
e
in
h
an
d
lin
g
im
b
ala
n
ce
d
class
if
icatio
n
task
s
.
I
t
en
h
an
ce
s
m
o
d
el
p
er
f
o
r
m
an
ce
b
y
u
tili
zin
g
a
co
n
tr
asti
v
e
lo
s
s
f
u
n
ctio
n
th
at
ac
ce
n
tu
ates th
e
d
is
ti
n
ctiv
en
ess
an
d
s
im
ilar
ities
b
etwe
en
s
am
p
les,
th
u
s
en
h
an
cin
g
th
e
m
o
d
el’
s
ca
p
ac
it
y
to
g
e
n
er
alize
ac
r
o
s
s
d
iv
er
s
e
d
ata
s
ce
n
ar
io
s
[
8
]
.
C
o
n
s
id
er
in
g
th
e
r
ec
en
t
im
p
r
o
v
em
en
ts
in
s
elf
-
s
u
p
er
v
is
ed
co
n
t
r
asti
v
e
lear
n
in
g
f
o
r
im
b
alan
ce
lear
n
in
g
,
w
e
ex
am
in
e
h
o
w
th
ese
em
er
g
in
g
ad
v
an
ce
m
en
ts
ca
n
en
h
an
ce
th
e
f
ield
o
f
im
b
alan
ce
class
if
icat
io
n
.
T
h
e
p
u
r
p
o
s
e
o
f
th
is
wo
r
k
is
to
ex
a
m
in
e
s
elf
-
s
u
p
er
v
is
ed
c
o
n
tr
asti
v
e
lear
n
in
g
ap
p
r
o
ac
h
es
th
at
h
av
e
b
ee
n
cr
ea
ted
in
th
e
p
ast f
iv
e
y
ea
r
s
to
ad
d
r
ess
th
e
is
s
u
e
o
f
im
b
alan
ce
d
class
if
icatio
n
.
T
h
e
aim
is
to
o
f
f
e
r
d
ir
e
ctio
n
to
ac
ad
em
ics
wh
o
ar
e
in
ter
ested
in
ex
p
lo
r
in
g
th
is
r
esear
ch
to
p
ic
i
n
th
e
co
n
tex
t o
f
class
im
b
alan
ce
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
is
p
ap
er
i
s
as
f
o
llo
ws
:
in
s
ec
t
io
n
2
,
we
p
r
o
v
id
e
a
co
n
cise
o
v
er
v
iew
o
f
th
e
p
r
o
m
in
e
n
t
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
m
eth
o
d
s
an
d
im
b
alan
ce
ap
p
r
o
ac
h
es
th
at
h
av
e
b
ee
n
d
ev
is
ed
.
T
h
en
,
we
p
r
o
v
i
d
e
th
e
m
eth
o
d
s
u
s
ed
in
co
n
d
u
ctin
g
th
is
r
ev
iew.
Sectio
n
4
p
r
o
v
id
es
an
ex
t
en
s
iv
e
ex
am
in
atio
n
o
f
th
e
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
a
s
tiv
e
lear
n
in
g
m
eth
o
d
s
th
at
h
av
e
b
ee
n
cr
ea
ted
to
a
d
d
r
es
s
th
e
ch
allen
g
e
o
f
im
b
alan
ce
d
ca
teg
o
r
izatio
n
.
Sectio
n
5
p
r
o
v
id
es
a
co
n
clu
s
io
n
to
th
e
r
e
p
o
r
t,
o
f
f
e
r
in
g
v
a
lu
ab
le
in
s
ig
h
ts
o
n
p
o
ten
tial f
u
tu
r
e
r
esear
ch
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
2
.
1
.
Self
-
s
up
er
v
is
e
d lea
rning
S
e
l
f
-
s
u
p
e
r
v
is
e
d
l
e
a
r
n
i
n
g
(
SS
L
)
h
a
s
g
a
r
n
e
r
e
d
s
i
g
n
i
f
i
c
a
n
t
i
n
t
e
r
es
t
l
a
t
el
y
a
s
a
v
ia
b
l
e
m
et
h
o
d
t
o
t
r
a
i
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
w
i
t
h
o
u
t
r
e
q
u
i
r
i
n
g
a
n
e
n
o
r
m
o
u
s
a
m
o
u
n
t
o
f
l
ab
e
l
l
e
d
d
a
t
a
.
SS
L
i
n
v
o
l
v
es
a
m
o
d
e
l
t
h
a
t
ca
n
p
r
e
d
i
ct
s
p
e
c
i
f
i
c
a
t
t
r
i
b
u
t
e
s
o
r
r
e
l
at
i
o
n
s
h
i
p
s
i
n
s
i
d
e
t
h
e
d
a
t
a
w
i
t
h
o
u
t
d
e
p
e
n
d
i
n
g
o
n
e
x
t
e
r
n
a
l
a
n
n
o
t
a
tio
n
s
.
T
h
i
s
e
n
a
b
l
es
t
h
e
m
o
d
e
l
t
o
a
c
q
u
i
r
e
v
a
l
u
a
b
l
e
r
e
p
r
e
s
e
n
t
at
i
o
n
s
t
h
at
c
a
n
b
e
r
e
f
i
n
e
d
f
o
r
p
a
r
t
i
c
u
l
a
r
d
u
t
i
es
i
n
t
h
e
f
u
t
u
r
e
[
9
]
,
[
1
0
]
.
2
.
1
.
1
.
P
rinciples
o
f
s
elf
-
s
up
er
v
is
ed
lea
rning
F
i
g
u
r
e
1
i
l
l
u
s
t
r
a
t
es
t
h
e
o
v
e
r
a
ll
p
r
o
c
e
s
s
o
f
s
el
f
-
s
u
p
e
r
v
is
e
d
le
a
r
n
i
n
g
,
w
h
e
r
e
r
e
p
r
e
s
e
n
ts
r
aw
d
a
t
a
,
r
e
p
r
e
s
e
n
t
s
l
a
b
e
ls
.
D
u
r
i
n
g
t
h
e
f
i
r
s
t
p
h
a
s
e
,
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
C
o
n
v
N
et
)
a
r
e
t
r
a
i
n
e
d
t
o
r
e
s
o
l
v
e
a
s
p
e
c
i
f
i
c
p
r
et
e
x
t
t
as
k
.
T
h
e
p
s
e
u
d
o
l
a
b
e
l
s
f
o
r
p
r
e
t
e
x
t
t
as
k
s
a
r
e
c
r
e
a
t
e
d
d
e
p
e
n
d
i
n
g
o
n
p
a
r
t
i
c
u
l
a
r
d
a
t
a
f
e
a
t
u
r
e
s
.
T
h
e
C
o
n
v
N
e
t
g
e
ts
t
r
a
i
n
e
d
t
o
a
c
q
u
i
r
e
k
n
o
w
l
e
d
g
e
o
f
e
n
t
i
t
y
c
h
a
r
a
c
t
e
r
i
s
t
i
cs
r
e
l
at
e
d
t
o
t
h
e
p
r
e
t
e
x
t
t
a
s
k
[
1
1
]
.
D
u
r
i
n
g
t
r
a
i
n
i
n
g
,
s
h
a
ll
o
w
e
r
b
l
o
c
k
s
o
f
t
h
e
C
o
n
v
N
e
t
p
r
i
m
a
r
i
l
y
d
e
t
e
ct
l
o
w
-
l
e
v
e
l
g
e
n
e
r
a
l
f
e
a
t
u
r
e
s
,
w
h
i
l
e
d
e
e
p
e
r
b
l
o
c
k
s
c
o
n
c
e
n
t
r
a
t
e
o
n
i
d
e
n
t
i
f
y
i
n
g
h
i
g
h
e
r
l
e
v
e
l
s
,
t
as
k
-
s
p
e
ci
f
i
c
c
h
a
r
a
ct
e
r
i
s
t
i
cs
l
i
k
e
o
b
j
e
c
ts
,
s
c
e
n
a
r
i
o
s
,
a
n
d
p
a
r
t
s
o
f
t
h
i
n
g
s
[
1
2
]
.
O
n
c
e
t
h
e
s
e
l
f
-
s
u
p
e
r
v
i
s
e
d
t
r
a
i
n
i
n
g
i
s
c
o
m
p
l
e
te
,
v
i
s
u
a
l
c
h
ar
a
c
t
e
r
is
t
i
cs
o
b
t
a
i
n
e
d
c
a
n
b
e
e
f
f
i
c
i
e
n
t
l
y
u
t
il
i
z
e
d
f
o
r
d
o
w
n
s
t
r
e
a
m
t
a
s
k
s
,
p
a
r
ti
c
u
l
a
r
l
y
i
n
c
as
e
s
w
it
h
s
p
a
r
s
e
d
a
t
a
.
P
r
e
-
t
r
a
i
n
e
d
m
o
d
e
ls
i
m
p
r
o
v
e
p
e
r
f
o
r
m
a
n
c
e
a
n
d
m
i
ti
g
a
t
e
t
h
e
i
s
s
u
e
o
f
o
v
e
r
-
f
i
t
t
i
n
g
.
O
n
l
y
t
h
e
v
i
s
u
a
l
c
h
a
r
a
c
t
e
r
i
s
ti
cs
f
r
o
m
t
h
e
i
n
i
t
i
a
l
l
a
y
e
r
s
a
r
e
c
o
n
v
e
y
e
d
f
o
r
t
r
a
i
n
i
n
g
t
h
e
s
u
p
e
r
v
i
s
e
d
d
o
w
n
s
t
r
ea
m
t
as
k
[
1
3
]
.
T
h
i
s
s
t
r
a
t
e
g
y
s
i
g
n
i
f
ica
n
t
l
y
e
n
h
a
n
c
es
d
o
w
n
s
t
r
e
a
m
t
a
s
k
s
'
p
e
r
f
o
r
m
a
n
c
e
c
o
m
p
a
r
e
d
t
o
t
r
a
i
n
i
n
g
t
h
e
m
o
d
e
l
s
f
r
o
m
t
h
e
b
e
g
i
n
n
i
n
g
[
1
4
]
,
[
1
5
]
.
H
u
a
n
g
e
t
a
l
.
[
1
6
]
c
a
t
e
g
o
r
i
e
s
s
e
l
f
-
s
u
p
e
r
v
is
e
d
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
i
n
t
o
t
h
r
ee
g
r
o
u
p
s
:
g
e
n
e
r
a
t
i
v
e
,
c
o
n
t
r
a
s
t
i
v
e
a
n
d
p
r
e
d
i
c
t
i
v
e
.
M
o
s
t
v
is
u
a
l
c
a
t
e
g
o
r
iz
a
t
i
o
n
p
r
o
b
l
e
m
s
b
e
n
e
f
i
t
f
r
o
m
c
o
n
t
r
a
s
ti
v
e
l
e
a
r
n
i
n
g
[
1
7
]
.
T
h
e
c
o
n
t
r
a
s
tiv
e
l
o
s
s
f
u
n
c
ti
o
n
i
n
c
r
e
as
e
s
t
h
e
s
i
m
i
l
a
r
it
y
o
f
i
m
a
g
e
s
t
h
a
t
b
e
l
o
n
g
t
o
t
h
e
s
a
m
e
o
b
j
e
c
t
c
l
a
s
s
.
I
t
r
e
d
u
c
e
s
t
h
e
s
i
m
il
a
r
i
t
y
b
e
t
w
e
e
n
t
h
o
s
e
t
h
a
t
b
e
l
o
n
g
t
o
d
i
s
t
i
n
ct
c
l
a
s
s
es
,
w
h
i
c
h
i
s
es
p
e
c
i
al
l
y
a
d
v
a
n
t
a
g
e
o
u
s
f
o
r
d
o
w
n
s
t
r
e
a
m
o
b
j
e
ct
d
e
t
e
ct
i
o
n
an
d
c
l
a
s
s
i
f
i
c
at
i
o
n
t
as
k
s
[
1
8
]
.
U
n
l
a
b
e
l
e
d
D
a
t
a
s
e
t
C
o
n
v
N
e
t
S
e
l
f
-
S
u
p
e
r
v
i
s
e
d
P
r
e
-
t
r
a
i
n
i
n
g
K
n
o
w
l
e
d
g
e
T
r
a
n
s
f
e
r
P
r
e
d
i
c
t
o
r
P
r
e
t
e
x
t
T
a
s
k
L
a
b
e
l
e
d
D
a
t
a
s
e
t
C
o
n
v
N
e
t
S
u
p
e
r
v
i
s
e
d
D
o
w
n
s
t
r
e
a
m
T
a
s
k
P
r
e
d
i
c
t
o
r
D
o
w
n
s
t
r
e
a
m
T
a
s
k
Fig
u
r
e
1
.
T
h
e
f
lo
w
o
f
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
2
.
1
.
2
.
Co
ntr
a
s
t
iv
e
lea
rning
f
ra
m
ewo
r
k
s
C
o
n
tr
asti
v
e
lear
n
in
g
is
a
n
e
w
p
o
p
u
lar
s
elf
-
s
u
p
er
v
is
ed
m
eth
o
d
an
d
h
as
d
em
o
n
s
tr
ated
ex
ce
p
tio
n
al
p
er
f
o
r
m
an
ce
ac
r
o
s
s
v
ar
io
u
s
co
m
p
u
ter
v
is
io
n
task
s
,
r
esu
ltin
g
in
wid
esp
r
ea
d
s
u
cc
ess
in
i
m
ag
e
[
1
9
]
,
[
2
0
]
,
tex
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
ev
is
itin
g
s
elf
-
s
u
p
ervi
s
ed
co
n
tr
a
s
tive
lea
r
n
in
g
fo
r
imb
a
la
n
ce
d
cla
s
s
ifica
tio
n
(
X
ia
o
lin
g
Ga
o
)
1951
[
2
1
]
,
[
2
2
]
,
au
d
i
o
[
2
3
]
,
[
2
4
]
,
an
d
v
id
eo
[
2
5
]
,
[
2
6
]
.
C
o
n
tr
a
s
tiv
e
lear
n
in
g
ex
tr
ac
ts
co
m
p
a
r
ab
le
o
r
d
is
s
im
ilar
r
ep
r
esen
tatio
n
s
f
r
o
m
d
ata
a
r
r
a
n
g
ed
in
p
air
s
b
ased
o
n
th
eir
s
i
m
ilar
ities
o
r
d
if
f
er
en
ce
s
.
A
n
a
p
p
lied
lo
s
s
f
u
n
ctio
n
f
o
r
co
n
tr
asti
v
e
lear
n
in
g
,
k
n
o
w
n
as I
n
f
o
NC
E
,
is
:
ℒ
,
+
,
{
−
}
=
−
l
og
e
xp
(
∙
+
/
)
e
xp
(
∙
+
/
)
+
∑
e
xp
(
∙
−
/
)
−
(
1
)
wh
er
e
r
ep
r
esen
ts
th
e
q
u
e
r
y
r
e
p
r
esen
tatio
n
,
+
is
th
e
r
ep
r
esen
ta
tio
n
o
f
p
o
s
itiv
e
k
e
y
s
am
p
les
a
n
d
−
d
en
o
tes
n
eg
ativ
e
k
ey
s
am
p
les,
with
τ
d
en
o
tin
g
a
tem
p
er
atu
r
e
h
y
p
er
-
p
ar
am
eter
.
E
q
u
atio
n
(
1
)
en
co
u
r
ag
es
th
e
m
o
d
el
to
clo
s
ely
alig
n
th
e
d
e
p
ictio
n
s
o
f
p
o
s
itiv
e
p
air
s
an
d
d
is
tan
ce
th
o
s
e
o
f
n
eg
ativ
e
p
air
s
.
C
o
n
tr
asti
v
e
m
eth
o
d
s
[
2
7
]
im
p
lem
en
t
th
e
a
p
p
r
o
ac
h
s
u
g
g
ested
in
r
ef
er
e
n
ce
[
2
8
]
th
at
d
en
s
e
co
n
tr
asti
v
e
lear
n
in
g
f
ac
ilit
ates
s
elf
-
s
u
p
er
v
is
ed
v
is
u
al
p
r
e
-
tr
ain
in
g
b
y
attr
ac
tin
g
p
o
s
itiv
e
s
am
p
le
p
ai
r
in
g
s
an
d
r
e
p
ellin
g
n
eg
ativ
e
o
n
es.
Mo
m
en
t
u
m
co
n
tr
ast
(
Mo
C
o
)
[
1
0
]
u
tili
ze
s
u
n
lab
eled
d
ata
to
cr
ea
te
p
r
e
-
tr
ai
n
ed
m
o
d
els
th
at
ca
n
b
e
f
u
r
th
er
r
ef
in
e
d
with
lab
elle
d
d
ata.
C
h
en
et
a
l.
[
9
]
u
tili
ze
d
SimCLR
to
attain
s
im
ilar
r
es
u
lts
as
a
s
u
p
er
v
is
ed
R
esNet
-
5
0
m
o
d
el
b
y
s
o
lely
tr
ain
in
g
a
lin
ea
r
class
if
ier
o
n
s
elf
-
s
u
p
er
v
is
ed
r
ep
r
esen
tatio
n
s
f
r
o
m
th
e
en
tire
I
m
ag
eNe
t
d
ataset.
C
h
en
et
a
l
.
[
2
0
]
u
p
g
r
ad
ed
Mo
C
o
to
M
o
C
o
v
2
,
en
a
b
lin
g
c
o
m
p
etitiv
e
r
esu
lts
with
s
h
o
r
t
b
atch
s
ize
tr
ain
in
g
o
n
th
e
e
n
tire
I
m
ag
eNe
t
[
2
9
]
.
B
YOL
[
3
0
]
an
d
SimSiam
[
3
1
]
aim
to
m
i
n
im
ize
th
e
d
is
tan
ce
b
etwe
en
p
air
s
o
f
p
o
s
itiv
e
s
am
p
les
an
d
an
asy
m
m
etr
ic
Siam
ese
n
etwo
r
k
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
p
o
p
u
lar
f
r
am
ewo
r
k
s
.
Fig
u
r
e
2
(
a)
d
ep
icts
th
e
Mo
C
o
f
r
am
ewo
r
k
,
Fig
u
r
e
2
(
b
)
s
h
o
ws
th
e
SimCL
R
f
r
am
ewo
r
k
,
an
d
Fig
u
r
e
2
(
c)
illu
s
tr
ates
th
e
B
YOL
f
r
am
ewo
r
k
.
C
o
n
t
r
asti
v
e
lear
n
in
g
f
r
am
ewo
r
k
s
aim
to
im
p
r
o
v
e
th
e
ag
r
ee
m
en
t
b
etwe
en
s
im
ilar
im
ag
es
wh
ile
d
is
tin
g
u
is
h
in
g
th
em
f
r
o
m
d
is
s
im
ilar
im
ag
es
u
s
in
g
a
co
n
tr
asti
v
e
lo
s
s
f
u
n
ctio
n
.
T
h
is
p
r
e
-
tr
ain
in
g
m
eth
o
d
f
o
r
ce
s
th
e
m
o
d
el
to
o
b
tain
ef
f
i
cien
t
r
ep
r
esen
tatio
n
s
.
Ap
p
r
o
a
ch
es
d
if
f
er
in
th
ei
r
tech
n
iq
u
es
f
o
r
g
e
n
er
atin
g
p
o
s
itiv
e
an
d
n
e
g
ativ
e
im
a
g
e
p
air
s
f
r
o
m
u
n
lab
eled
d
ata
an
d
d
a
ta
s
elec
tio
n
in
p
r
e
-
tr
ain
in
g
.
W
an
g
et
a
l.
[
3
2
]
an
a
ly
ze
co
n
tr
asti
v
e
lear
n
in
g
in
te
r
m
s
o
f
th
e
r
eg
u
lar
ity
a
n
d
alig
n
m
en
t
o
f
ac
q
u
ir
ed
r
ep
r
esen
tatio
n
s
.
Ko
tar
et
a
l.
[
3
3
]
co
m
p
r
eh
en
s
iv
ely
an
al
y
ze
co
n
tr
asti
v
e
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
tech
n
i
q
u
es.
I
t
ex
p
lo
r
es
th
e
e
f
f
ec
ts
o
f
d
if
f
er
e
n
t
tr
ain
in
g
s
tr
ateg
ies
an
d
d
ata
s
ets
o
n
p
er
f
o
r
m
an
ce
i
n
v
a
r
io
u
s
d
o
wn
s
tr
ea
m
task
s
,
co
n
clu
d
in
g
th
at
th
ese
ap
p
r
o
ac
h
es si
g
n
if
ican
tly
ad
v
a
n
ce
s
tate
-
of
-
th
e
-
ar
t r
ep
r
esen
tatio
n
lear
n
in
g
.
E
n
c
o
d
e
r
M
o
m
e
n
t
u
m
E
n
c
o
d
e
r
I
n
f
o
N
C
E
S
i
m
i
l
a
r
i
t
y
E
n
c
o
d
e
r
E
n
c
o
d
e
r
M
a
x
i
m
i
z
e
S
i
m
i
l
a
r
i
t
y
P
r
o
j
e
c
t
i
o
n
H
e
a
d
P
r
o
j
e
c
t
i
o
n
H
e
a
d
M
e
a
n
S
q
u
a
r
e
L
o
s
s
S
t
o
p
-
g
r
a
d
E
n
c
o
d
e
r
E
n
c
o
d
e
r
P
r
o
j
e
c
t
i
o
n
H
e
a
d
P
r
o
j
e
c
t
i
o
n
H
e
a
d
P
r
e
d
i
c
t
o
r
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
I
ll
u
s
tr
atio
n
o
f
t
h
e
co
n
tr
asti
v
e
lear
n
in
g
f
r
am
ewo
r
k
s
(
a)
Mo
C
o
,
(
b
)
SimCLR,
an
d
(
c)
B
YOL
2
.
2
.
I
m
ba
la
nce
lea
rning
2
.
2
.
1
.
Da
t
a
-
lev
el
Data
-
lev
el
tech
n
iq
u
es
en
co
m
p
ass
th
e
u
tili
za
tio
n
o
f
o
v
er
-
s
am
p
lin
g
an
d
u
n
d
er
-
s
am
p
lin
g
.
T
h
e
d
ata
-
lev
el
tech
n
iq
u
e
in
v
o
lv
es
alter
i
n
g
th
e
tr
ain
in
g
d
ata
to
ac
h
iev
e
an
e
q
u
itab
le
d
is
tr
ib
u
tio
n
o
f
class
es.
T
h
e
u
n
d
er
-
s
am
p
lin
g
m
eth
o
d
b
alan
ce
s
th
e
d
ata
b
y
d
eletin
g
s
am
p
les o
f
th
e
m
ajo
r
ity
class
,
wh
ich
m
ay
r
e
d
u
ce
s
o
m
e
h
elp
f
u
l
in
f
o
r
m
atio
n
in
th
e
d
atasets
.
On
th
e
co
n
tr
ar
y
,
th
e
o
v
er
-
s
am
p
l
in
g
m
eth
o
d
b
alan
ce
s
th
e
d
ata
b
y
au
g
m
e
n
tin
g
th
e
m
in
o
r
ity
class
with
ad
d
itio
n
al
s
am
p
les
b
y
r
ep
ea
tin
g
o
r
g
en
e
r
atin
g
n
ew
in
s
tan
ce
s
,
wh
ich
ca
u
s
es
th
e
lear
n
er
to
o
v
er
f
it.
C
h
awla
et
a
l.
[
3
4
]
cr
e
ated
s
y
n
th
etic
m
in
o
r
ity
o
v
er
-
s
am
p
lin
g
tec
h
n
iq
u
e
(
SMOT
E
)
to
o
v
e
r
co
m
e
th
ese
co
n
ce
r
n
s
b
y
g
e
n
er
atin
g
n
ew
in
s
tan
ce
s
o
f
th
e
m
i
n
o
r
ity
cla
s
s
b
ased
o
n
th
e
k
-
n
ea
r
est
n
e
ig
h
b
o
r
s
.
W
h
en
th
e
m
in
o
r
ity
class
co
n
s
is
ts
o
f
n
u
m
er
o
u
s
s
m
all
s
ep
ar
ate
clu
s
ter
s
,
u
s
in
g
SMOT
E
ca
n
lead
to
m
o
r
e
class
o
v
er
lap
an
d
r
aise
th
e
co
m
p
lex
ity
o
f
th
e
c
lass
if
icatio
n
task
[
3
5
]
.
Var
io
u
s
s
o
lu
tio
n
s
h
av
e
b
ee
n
s
u
g
g
e
s
ted
to
tack
le
th
ese
s
h
o
r
tco
m
in
g
s
b
y
eith
er
i
n
co
r
p
o
r
atin
g
b
o
th
class
es d
u
r
in
g
g
e
n
er
atio
n
o
r
as a
s
u
b
s
eq
u
e
n
t c
lean
in
g
p
r
o
ce
s
s
.
T
h
e
r
e
s
e
a
r
c
h
e
r
p
r
o
p
o
s
e
d
B
o
r
d
e
r
l
i
n
e
-
S
MO
T
E
[
3
6
]
,
w
h
ic
h
o
n
l
y
o
v
e
r
s
a
m
p
l
es
o
r
s
t
r
e
n
g
t
h
e
n
s
t
h
e
b
o
r
d
e
r
l
i
n
e
m
i
n
o
r
i
t
y
s
a
m
p
l
es
.
A
d
a
p
t
i
v
e
S
y
n
t
h
e
ti
c
s
a
m
p
l
i
n
g
a
p
p
r
o
a
c
h
(
A
D
A
S
YN
)
[
3
7
]
u
t
i
l
i
z
es
a
w
ei
g
h
t
e
d
d
i
s
t
r
i
b
u
ti
o
n
t
o
a
l
l
o
c
a
te
v
a
r
i
o
u
s
m
i
n
o
r
i
t
y
s
a
m
p
l
es
b
a
s
e
d
o
n
t
h
e
l
e
v
e
l
o
f
l
e
a
r
n
i
n
g
c
o
m
p
l
e
x
i
t
y
.
S
a
f
e
-
l
e
v
e
l
-
S
M
O
T
E
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
4
9
-
1
9
6
0
1952
[
3
8
]
a
n
d
n
o
i
s
e
r
e
d
u
c
ti
o
n
a
p
r
io
r
i
s
y
n
t
h
e
ti
c
o
v
e
r
-
s
a
m
p
l
i
n
g
tec
h
n
i
q
u
e
(
N
R
A
S
)
[
3
9
]
a
r
e
m
e
an
t
t
o
m
i
n
i
m
i
z
e
t
h
e
c
h
a
n
c
e
o
f
i
n
t
r
o
d
u
c
i
n
g
d
i
s
r
u
p
t
i
v
e
a
r
t
i
f
i
c
i
al
d
a
t
a
p
o
i
n
t
s
w
it
h
i
n
t
h
e
m
a
i
n
c
l
a
s
s
a
r
e
a
.
T
h
e
s
a
m
p
l
i
n
g
w
i
t
h
t
h
e
m
a
j
o
r
i
t
y
(
S
W
I
M
)
[
4
0
]
a
p
p
r
o
a
c
h
u
s
e
s
M
a
h
a
l
a
n
o
b
i
s
d
is
t
a
n
c
e
t
o
l
o
c
a
te
s
y
n
t
h
e
t
i
c
s
a
m
p
l
es
b
a
s
e
d
o
n
b
o
t
h
c
l
a
s
s
e
s
'
s
a
m
p
l
es
.
R
a
d
i
al
-
b
a
s
e
d
o
v
e
r
s
a
m
p
l
i
n
g
(
R
B
O
)
[
4
1
]
g
e
n
e
r
at
es
m
i
n
o
r
i
t
y
o
b
j
e
c
t
s
u
s
i
n
g
r
a
d
i
a
l
b
a
s
i
s
f
u
n
c
t
i
o
n
s
a
n
d
p
o
t
e
n
t
ia
l
e
s
t
i
m
a
t
i
o
n
.
C
o
m
b
i
n
e
d
c
l
e
a
n
i
n
g
a
n
d
r
e
s
a
m
p
l
i
n
g
(
C
C
R
)
[
4
2
]
m
e
t
h
o
d
c
l
e
a
n
s
m
i
n
o
r
i
t
y
o
b
j
e
c
t
d
e
c
i
s
i
o
n
b
o
r
d
e
r
s
a
n
d
g
u
i
d
e
s
s
y
n
t
h
e
t
i
c
o
v
e
r
s
a
m
p
l
i
n
g
.
S
e
v
e
r
a
l
w
o
r
k
s
h
a
v
e
e
x
p
l
o
r
e
d
h
o
w
t
o
i
m
p
r
o
v
e
t
h
e
s
y
n
th
e
t
i
c
o
v
e
r
s
a
m
p
l
i
n
g
m
e
t
h
o
d
t
o
b
e
s
u
i
t
a
b
l
e
f
o
r
m
u
l
t
i
-
c
l
a
s
s
i
m
b
a
l
a
n
c
e
is
s
u
e
s
.
M
a
h
ala
n
o
b
i
s
d
i
s
t
a
n
ce
-
b
a
s
e
d
o
v
e
r
-
s
am
p
l
i
n
g
(
M
D
O
)
[
4
3
]
u
s
e
d
t
h
e
s
a
m
e
M
a
h
a
l
a
n
o
b
is
d
is
t
a
n
c
e
as
t
h
e
cl
as
s
m
ea
n
o
f
e
a
ch
i
n
s
p
e
c
ti
o
n
.
Z
h
u
e
t
a
l
.
[
4
4
]
p
r
o
p
o
s
e
d
k
-
NN
-
b
a
s
e
d
s
y
n
t
h
e
t
i
c
m
i
n
o
r
it
y
o
v
e
r
s
a
m
p
l
in
g
(
S
M
O
M
)
,
w
h
i
c
h
b
a
l
a
n
c
es
m
i
n
o
r
i
t
y
s
a
m
p
l
e
d
i
r
e
ct
i
o
n
t
o
b
u
i
l
d
n
e
w
p
r
o
t
o
t
y
p
e
s
.
S
y
n
t
h
e
t
i
c
o
v
e
r
s
a
m
p
l
i
n
g
w
i
t
h
th
e
m
i
n
o
r
i
t
y
a
n
d
m
a
j
o
r
i
t
y
c
l
ass
e
s
(
S
OM
M
)
[
4
5
]
u
s
e
s
y
n
t
h
e
ti
c
o
v
e
r
s
a
m
p
li
n
g
t
o
c
r
e
a
t
e
s
y
n
t
h
et
i
c
s
a
m
p
l
es
f
o
r
m
in
o
r
i
t
y
a
n
d
m
a
j
o
r
i
t
y
c
l
as
s
e
s
.
2
.
2
.
2
.
Alg
o
rit
hm
-
lev
el
T
h
e
i
m
b
a
l
a
n
c
e
p
r
o
b
l
e
m
i
s
s
o
lv
e
d
a
t
t
h
e
al
g
o
r
i
t
h
m
le
v
e
l
,
u
til
i
z
i
n
g
v
a
r
y
i
n
g
m
is
c
la
s
s
i
f
i
ca
t
i
o
n
c
o
s
ts
t
o
m
a
k
e
c
l
as
s
i
f
i
e
r
s
p
r
i
o
r
it
i
z
e
t
h
e
m
i
n
o
r
i
t
y
c
l
ass
[
4
6
]
.
I
n
t
h
e
i
m
b
a
l
a
n
c
e
i
s
s
u
e
,
a
f
a
ls
e
n
e
g
at
i
v
e
p
r
e
d
i
c
t
i
o
n
s
h
o
u
l
d
c
o
s
t
m
o
r
e
t
h
a
n
a
f
a
l
s
e
p
o
s
i
t
i
v
e
i
f
th
e
m
i
n
o
r
i
t
y
c
l
a
s
s
is
p
o
s
it
i
v
e
in
c
l
a
s
s
i
f
i
c
at
i
o
n
o
u
t
c
o
m
e
s
.
C
las
s
i
f
i
c
a
ti
o
n
a
l
m
e
t
h
o
d
s
c
a
n
i
n
c
l
u
d
e
t
h
es
e
e
x
p
e
n
s
es
d
u
r
i
n
g
m
o
d
e
l
t
r
a
i
n
i
n
g
t
o
r
e
d
u
c
e
im
b
a
l
a
n
c
e
d
d
a
t
a
.
T
h
e
c
l
ass
i
c
Ad
a
-
b
o
o
s
t
a
l
g
o
r
i
t
h
m
r
e
d
u
c
e
s
c
l
a
s
s
i
f
i
e
r
g
e
n
e
r
a
t
i
o
n
e
r
r
o
r
.
S
u
n
e
t
a
l
.
[
4
7
]
s
t
u
d
i
e
d
m
e
t
a
-
t
e
c
h
n
i
q
u
e
s
f
o
r
u
n
b
a
l
a
n
c
e
d
d
a
t
a
.
T
h
e
y
c
o
m
b
i
n
e
d
c
o
s
t
-
s
e
n
s
it
i
v
e
l
ea
r
n
i
n
g
w
i
t
h
Ad
a
B
o
o
s
t
t
o
d
e
v
e
l
o
p
t
h
r
e
e
c
o
s
t
-
s
e
n
s
it
i
v
e
b
o
o
s
ti
n
g
m
e
t
h
o
d
s
to
i
m
p
r
o
v
e
p
o
s
i
t
i
v
e
c
l
a
s
s
c
a
t
e
g
o
r
i
z
at
i
o
n
.
B
es
i
d
es
,
L
i
n
e
t
a
l
.
[
4
8
]
p
r
o
p
o
s
e
d
t
h
e
l
o
s
s
f
u
n
c
t
i
o
n
o
f
F
o
c
a
l
L
o
s
s
,
i
n
w
h
i
c
h
a
p
e
n
a
l
t
y
i
s
a
p
p
l
i
e
d
f
o
r
e
a
c
h
c
a
t
e
g
o
r
y
u
t
il
i
z
i
n
g
a
c
o
s
t
m
a
t
r
i
x
.
I
n
c
r
e
a
s
e
t
h
e
w
ei
g
h
t
o
f
t
h
e
m
i
n
o
r
i
t
y
c
l
a
s
s
t
o
r
e
d
u
c
e
t
h
e
p
o
s
s
i
b
il
i
t
y
o
f
t
h
e
cl
a
s
s
b
ei
n
g
m
i
s
c
la
s
s
i
f
i
e
d
.
C
l
a
s
s
-
b
a
l
a
n
ce
d
(
CB
)
f
o
c
a
l
l
o
s
s
[
4
9
]
a
d
d
s
a
cl
a
s
s
-
b
a
l
a
n
c
e
d
f
a
ct
o
r
f
o
r
c
l
a
s
s
d
is
p
e
r
s
i
o
n
.
T
h
i
s
r
e
c
a
li
b
r
at
i
o
n
e
n
s
u
r
es
t
h
at
t
h
e
m
o
d
el
p
r
io
r
i
t
i
z
es
c
la
s
s
e
s
b
a
s
e
d
o
n
t
h
e
i
r
d
a
t
a
r
e
p
r
e
s
e
n
t
a
t
i
o
n
,
l
o
w
e
r
i
n
g
t
h
e
i
n
f
l
u
e
n
c
e
o
f
t
h
e
m
a
j
o
r
i
t
y
cl
as
s
.
p
r
o
g
r
es
s
i
v
e
m
a
r
g
i
n
l
o
s
s
(
PM
L
)
[
5
0
]
w
eig
h
t
s
d
e
c
is
i
o
n
b
o
r
d
e
r
s
a
m
p
l
es
b
e
c
a
u
s
e
t
h
e
y
d
e
f
i
n
e
c
la
s
s
s
e
p
a
r
a
t
i
o
n
s
.
L
o
n
g
-
t
a
il
e
d
m
u
l
t
i
-
l
a
b
el
d
a
t
as
et
s
a
r
e
c
o
m
p
l
e
x
e
n
o
u
g
h
t
o
o
p
t
i
m
i
ze
w
i
t
h
s
i
n
g
l
e
-
l
a
b
el
as
s
u
m
p
t
i
o
n
s
.
T
h
e
e
m
p
i
r
i
c
a
l
f
i
n
d
i
n
g
s
il
l
u
s
t
r
a
te
t
h
a
t
t
h
e
i
m
p
r
o
v
e
m
e
n
t
o
f
t
h
e
p
r
e
c
i
s
i
o
n
o
f
t
h
e
l
o
s
s
f
u
n
c
t
i
o
n
v
a
r
i
e
s
wi
t
h
d
i
f
f
e
r
e
n
t
d
a
t
a
s
e
ts
.
Al
g
o
r
i
t
h
m
-
l
e
v
e
l
a
p
p
r
o
a
c
h
e
s
l
ac
k
f
l
e
x
i
b
il
i
t
y
c
o
m
p
a
r
e
d
t
o
d
a
t
a
-
l
e
v
e
l
a
l
t
e
r
n
a
ti
v
e
s
[
5
1
]
.
2
.
2
.
3
.
E
ns
em
ble le
a
rning
E
n
s
em
b
le
lear
n
in
g
e
n
h
an
ce
s
p
r
ed
icted
ac
cu
r
ac
y
b
y
co
m
b
in
in
g
p
r
ed
ictio
n
s
f
r
o
m
m
an
y
m
o
d
els.
E
n
s
em
b
le
lear
n
in
g
in
clu
d
es
b
ag
g
in
g
,
b
o
o
s
tin
g
,
an
d
s
tack
in
g
m
eth
o
d
s
[
5
2
]
.
E
n
s
em
b
le
m
e
th
o
d
s
ar
e
f
r
eq
u
en
tly
em
p
lo
y
ed
to
ad
d
r
ess
th
e
is
s
u
e
o
f
class
im
b
alan
ce
.
Fo
r
e
x
am
p
le,
C
h
awla
et
a
l.
[
5
3
]
p
r
o
p
o
s
e
d
SMOT
E
B
ag
g
in
g
,
wh
ich
u
s
es
b
ag
g
in
g
an
d
SMO
T
E
to
b
u
ild
m
u
lti
-
class
if
ier
s
t
o
d
iv
er
s
if
y
f
a
k
e
s
am
p
les.
SMOT
E
B
o
o
s
t
g
en
er
ates
s
y
n
th
etic
m
in
o
r
ity
class
s
am
p
les
th
r
o
u
g
h
o
u
t
ea
c
h
b
o
o
s
tin
g
it
er
atio
n
u
s
in
g
SMOT
E
an
d
a
b
o
o
s
tin
g
tech
n
i
q
u
e.
Seif
f
er
t
[
5
4
]
in
tr
o
d
u
ce
d
R
USB
o
o
s
t,
a
m
eth
o
d
t
h
at
u
tili
ze
s
r
an
d
o
m
u
n
d
er
s
am
p
lin
g
.
R
USB
o
o
s
t
ca
n
d
ec
r
ea
s
e
tr
ain
in
g
tim
e
wh
ile
u
tili
zin
g
Ad
aBo
o
s
t
to
en
h
an
ce
p
er
f
o
r
m
an
ce
.
L
v
et
a
l.
[
5
5
]
im
p
l
em
en
ted
th
e
o
v
e
r
-
s
am
p
lin
g
SMOT
E
an
d
Ad
aB
o
o
s
t
alg
o
r
ith
m
to
b
alan
ce
cr
e
d
it
ca
r
d
co
n
s
u
m
p
tio
n
d
ata.
E
v
id
en
ce
s
h
o
ws
th
at
SMOT
E
-
Ad
aBo
o
s
t
ex
ce
ed
s
Ad
aBo
o
s
t.
I
leb
er
i
et
a
l.
[
5
6
]
s
u
g
g
ested
a
m
ac
h
i
n
e
lear
n
in
g
ap
p
r
o
ac
h
f
o
r
id
en
tify
in
g
i
n
s
tan
ce
s
o
f
cr
ed
i
t
ca
r
d
f
r
au
d
,
an
d
th
e
d
ataset
was
r
eb
alan
ce
d
u
s
in
g
SMO
T
E
.
Su
et
a
l.
[
5
7
]
p
r
o
p
o
s
ed
a
m
o
d
el
th
at
u
tili
ze
s
SMOT
E
-
Ad
aBo
o
s
t.
T
h
e
r
es
u
lts
d
em
o
n
s
tr
ate
en
h
an
ce
d
id
en
tific
atio
n
o
f
th
e
in
ten
d
ed
o
b
jectiv
e
o
f
th
e
c
o
m
b
at
tar
g
et
in
th
e
p
r
esen
ce
o
f
d
is
p
r
o
p
o
r
tio
n
ate
d
ata.
E
d
war
d
et
a
l.
[
5
8
]
p
r
esen
t
a
n
o
v
el
r
eb
alan
cin
g
f
r
am
ewo
r
k
,
in
co
r
p
o
r
atin
g
SMOT
E
a
n
d
clu
s
ter
-
b
ased
u
n
d
er
s
am
p
lin
g
tech
n
iq
u
e
(
SC
UT
)
,
an
d
r
ec
u
r
s
iv
e
f
ea
tu
r
e
elim
in
ati
o
n
(
R
FE)
f
o
r
im
p
r
o
v
ed
m
u
lti
-
class
clas
s
if
icatio
n
p
er
f
o
r
m
an
ce
in
ad
d
r
ess
in
g
th
e
ch
allen
g
es
o
f
im
b
alan
ce
d
m
e
d
ical
d
atasets
.
Gao
et
a
l.
[
5
9
]
ex
p
lo
r
e
t
h
e
ef
f
ec
tiv
e
n
ess
o
f
co
m
b
in
in
g
SMOT
E
with
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
m
o
d
els
an
d
th
e
b
o
o
s
tin
g
m
eth
o
d
to
ad
d
r
ess
im
b
alan
ce
d
im
ag
e
class
if
icatio
n
task
s
.
T
h
e
b
o
o
s
tin
g
en
s
em
b
le
tech
n
iq
u
e
is
g
en
er
ally
m
o
r
e
e
f
f
ec
tiv
e
t
h
an
u
s
in
g
a
s
in
g
le
class
if
ier
to
ad
d
r
ess
th
e
is
s
u
e
o
f
class
im
b
alan
ce
.
I
t sh
o
ws s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
r
eso
lv
i
n
g
th
is
is
s
u
e.
3.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
e
r
esear
ch
tech
n
iq
u
es
p
ar
t
p
r
im
ar
ily
f
o
c
u
s
es
o
n
th
e
p
lan
n
in
g
,
ex
ec
u
tio
n
,
an
d
p
r
esen
tatio
n
o
f
th
e
r
ev
iew
f
in
d
in
g
s
.
I
n
itially
,
th
e
r
elev
an
t
r
esear
ch
q
u
esti
o
n
s
o
n
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
le
ar
n
in
g
m
eth
o
d
s
f
o
r
h
an
d
lin
g
im
b
alan
ce
d
d
ata
ar
e
f
o
r
m
u
late
d
an
d
d
ef
in
e
d
.
Sec
o
n
d
,
th
e
r
elev
an
t
liter
atu
r
e
an
d
r
elate
d
f
ac
ts
ar
e
ex
tr
ac
ted
b
y
s
ea
r
ch
in
g
v
ar
io
u
s
d
atab
ases
.
Fin
ally
,
a
s
y
s
tem
atic
r
ev
iew
o
f
th
e
r
esu
lts
r
ep
o
r
t i
s
wr
itten
.
3
.
1
.
Resea
rc
h que
s
t
io
ns
T
h
e
in
itial
s
tag
e
i
n
ca
r
r
y
in
g
o
u
t
a
s
y
s
tem
atic
r
e
v
iew
in
v
o
lv
es
id
en
tify
in
g
th
e
r
esear
ch
q
u
esti
o
n
.
T
h
is
p
h
ase
s
h
o
u
ld
b
e
co
n
cise a
n
d
s
tr
aig
h
tf
o
r
wa
r
d
.
T
h
ese
ar
e
th
e
r
esear
ch
in
q
u
ir
ies with
in
th
e
s
c
o
p
e
o
f
th
is
s
tu
d
y
:
Q1
: Wh
at
is
th
e
p
r
esen
t statu
s
o
f
r
esear
ch
o
n
s
elf
-
s
u
p
e
r
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
f
o
r
im
b
al
an
ce
d
d
ata?
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
ev
is
itin
g
s
elf
-
s
u
p
ervi
s
ed
co
n
tr
a
s
tive
lea
r
n
in
g
fo
r
imb
a
la
n
ce
d
cla
s
s
ifica
tio
n
(
X
ia
o
lin
g
Ga
o
)
1953
Q2
:
W
h
at
is
th
e
m
o
s
t
ef
f
ec
ti
v
e
ap
p
r
o
ac
h
u
s
in
g
s
elf
-
s
u
p
e
r
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
to
ad
d
r
ess
im
b
alan
ce
d
d
ata
class
if
icatio
n
?
Q3
: Wh
at
ar
e
th
e
m
o
s
t c
r
itical
g
ap
s
an
d
s
h
o
r
tco
m
in
g
s
in
th
e
r
ev
iewe
d
r
esear
ch
?
3
.
2
.
Sea
rc
h
s
t
ra
t
eg
y
T
h
e
s
ea
r
ch
was
co
n
d
u
cted
u
s
in
g
s
p
ec
if
ic
ter
m
s
,
s
u
ch
as
"I
m
b
alan
ce
"
co
m
b
in
ed
with
t
h
e
"AN
D"
o
p
er
ato
r
an
d
"c
o
n
tr
asti
v
e
lea
r
n
in
g
"
alo
n
g
with
v
ar
io
u
s
s
y
n
o
n
y
m
s
,
as
in
d
icate
d
in
T
ab
l
e
1
.
T
o
en
s
u
r
e
th
e
s
u
r
v
ey
in
clu
d
e
d
o
n
ly
r
elev
a
n
t
s
cien
tific
wo
r
k
s
,
ad
d
itio
n
al
m
o
d
if
icatio
n
s
wer
e
im
p
lem
en
ted
in
ea
ch
s
ea
r
ch
en
g
in
e.
T
h
ese
m
o
d
if
icatio
n
s
ex
clu
d
ed
a
n
y
p
u
b
licatio
n
s
o
t
h
er
th
an
jo
u
r
n
al
an
d
co
n
f
er
e
n
ce
p
ap
er
s
,
th
er
e
b
y
r
ef
in
in
g
t
h
e
s
ea
r
ch
r
esu
lts
.
3
.
3
.
C
rit
er
ia
f
o
r
i
nclus
io
n a
nd
ex
clus
io
n
C
r
iter
ia
f
o
r
in
clu
s
io
n
wer
e
e
s
tab
lis
h
ed
to
ca
teg
o
r
ize
a
r
ticles
r
etr
iev
ed
f
r
o
m
s
cien
tific
d
atab
ases
,
en
s
u
r
in
g
th
e
co
llectio
n
o
f
p
e
r
tin
en
t
in
f
o
r
m
atio
n
r
elate
d
to
t
h
e
r
esear
ch
in
q
u
ir
ies.
On
ly
d
o
cu
m
en
ts
th
at
m
et
th
ese
s
p
ec
if
ic
cr
iter
ia,
as
o
u
tli
n
ed
in
T
ab
le
2
,
wer
e
co
n
s
id
er
ed
f
o
r
f
u
r
th
e
r
a
n
aly
s
is
.
T
h
is
a
p
p
r
o
ac
h
h
elp
ed
to
f
o
cu
s
th
e
r
e
v
iew
o
n
r
elev
an
t p
u
b
licatio
n
s
th
at
d
ir
ec
tly
a
d
d
r
e
s
s
ed
th
e
r
esear
ch
q
u
esti
o
n
s
.
3
.
4
.
Co
nd
uct
ing
re
v
iew
pro
ce
s
s
T
h
i
s
s
e
ct
i
o
n
d
e
t
ai
ls
t
h
e
p
r
a
c
ti
ca
l
e
x
e
c
u
ti
o
n
o
f
t
h
e
r
e
v
i
ew
d
e
p
ic
t
e
d
i
n
F
i
g
u
r
e
3
.
T
h
e
p
r
o
c
ess
en
t
a
i
l
e
d
t
h
e
i
d
e
n
t
i
f
i
c
at
i
o
n
,
s
c
r
e
e
n
i
n
g
,
a
s
s
ess
m
e
n
t
o
f
e
l
i
g
i
b
i
l
it
y
a
n
d
i
n
c
l
u
s
io
n
.
T
h
i
s
s
t
r
a
i
g
h
t
f
o
r
w
a
r
d
g
r
a
p
h
i
c
c
l
e
a
r
l
y
s
h
o
ws
t
h
e
m
e
t
h
o
d
i
c
a
l
a
p
p
r
o
a
c
h
t
o
c
h
o
o
s
i
n
g
r
e
l
e
v
a
n
t
s
t
u
d
i
es
f
o
r
r
e
v
iew
.
A
t
i
m
e
f
r
a
m
e
o
f
u
p
t
o
f
i
v
e
y
e
a
r
s
w
a
s
s
e
t
t
o
c
a
p
i
t
a
li
z
e
o
n
n
e
w
r
e
s
e
a
r
c
h
f
i
n
d
i
n
g
s
a
n
d
i
n
c
o
r
p
o
r
a
t
e
a
d
d
i
t
i
o
n
a
l
u
s
e
f
u
l
i
n
f
o
r
m
a
t
i
o
n
i
n
t
o
t
h
e
s
t
u
d
y
.
I
n
i
t
i
a
ll
y
,
t
h
e
8
4
2
p
a
p
e
r
s
o
b
t
a
i
n
e
d
f
r
o
m
t
h
e
A
C
M
Di
g
i
t
al
L
i
b
r
a
r
y
,
I
E
E
E
E
x
p
l
o
r
e
,
S
c
i
e
n
c
e
D
i
r
e
c
t
,
S
p
r
i
n
g
er
L
i
n
k
,
a
n
d
S
c
o
p
u
s
d
a
t
a
b
a
s
es
w
e
r
e
r
e
f
i
n
e
d
b
a
s
e
d
o
n
t
h
e
s
p
e
c
i
f
i
e
d
c
r
i
t
e
r
i
a
f
o
r
i
n
c
l
u
s
i
o
n
a
n
d
e
x
c
l
u
s
i
o
n
.
U
lt
i
m
at
e
l
y
,
a
t
o
t
a
l
o
f
7
9
8
a
r
t
i
c
l
es
w
e
r
e
d
e
e
m
e
d
i
n
e
li
g
i
b
l
e
a
n
d
e
x
c
l
u
d
e
d
,
w
h
i
l
e
4
4
a
r
t
i
cl
es
m
e
t
t
h
e
c
r
i
t
e
r
ia
a
n
d
w
e
r
e
c
o
n
s
i
d
e
r
e
d
a
d
m
is
s
i
b
l
e
.
T
ab
le
1
.
Qu
e
r
y
f
o
r
s
ea
r
ch
Mai
n
s
e
arc
h
s
t
ri
n
g
("
Im
b
a
l
a
n
ce
"
o
r
"
u
n
b
al
a
n
c
e"
o
r
"
s
k
e
w
"
)
an
d
(
"
c
l
a
s
s
i
f
i
ca
t
i
o
n
"
o
r
"
rec
o
g
n
i
t
i
o
n
"
)
a
n
d
("
Se
l
f
-
s
u
p
er
v
i
s
e
d
"
o
r
"
U
n
s
u
p
erv
i
s
e
d
"
)
a
n
d
("
c
o
n
t
r
as
t
i
v
e
l
ear
n
i
n
g
"
o
r
"
c
o
n
t
r
as
t
i
v
e
met
h
o
d
"
o
r
"
c
o
n
t
ra
s
t
i
v
e
t
ec
h
n
i
q
u
e
"
)
T
ab
le
2
.
T
h
e
cr
iter
io
n
f
o
r
s
elec
tio
n
In
c
l
u
s
i
o
n
E
x
cl
u
s
i
o
n
Cri
t
er
i
o
n
E
n
g
l
i
s
h
art
i
c
l
e
N
o
n
-
E
n
g
l
i
s
h
ar
t
i
c
l
e
W
h
e
t
h
er
t
h
e
l
i
t
e
rat
u
r
e
i
s
i
n
E
n
g
l
i
s
h
S
i
n
ce
2
0
19
Bef
o
re
2
0
1
9
A
t
i
me
l
i
n
e
J
o
u
r
n
a
l
art
i
c
l
e
a
n
d
co
n
fer
en
ce
p
r
o
ce
ed
i
n
g
Bo
o
k
a
n
d
re
v
i
e
w
G
en
re
o
f
l
i
t
er
at
u
re
S
t
a
r
t
C
h
o
o
s
e
d
i
g
i
t
a
l
l
i
b
r
a
r
i
e
s
S
e
a
r
c
h
s
t
r
a
t
e
g
y
I
n
i
t
i
a
t
e
p
i
l
o
t
s
e
a
r
c
h
M
o
s
t
k
n
o
w
n
s
t
u
d
i
e
s
f
o
u
n
d
?
S
e
a
r
c
h
s
t
r
i
n
g
r
e
f
i
n
e
m
e
n
t
R
e
t
r
i
e
v
e
i
n
i
t
i
a
l
p
r
i
m
a
r
y
s
t
u
d
y
l
i
s
t
(
8
4
2
)
D
i
g
i
t
a
l
l
i
b
r
a
r
i
e
s
·
A
C
M
D
i
g
i
t
a
l
L
i
b
r
a
r
y
(
8
5
)
·
I
E
E
E
E
x
p
l
o
r
e
(
1
8
6
)
·
S
c
i
e
n
c
e
D
i
r
e
c
t
(
4
9
3
)
·
S
p
r
i
n
g
e
r
L
i
n
k
(
2
0
)
·
S
c
o
p
u
s
(
5
8
)
B
a
s
e
d
o
n
t
i
t
l
e
a
n
d
a
b
s
t
r
a
c
t
,
e
x
c
l
u
d
i
n
g
p
r
i
m
a
r
y
r
e
s
e
a
r
c
h
(
1
2
6
)
F
u
l
l
-
t
e
x
t
p
r
i
m
a
r
y
s
t
u
d
i
e
s
a
r
e
e
x
c
l
u
d
e
d
(
4
4
)
F
i
n
i
s
h
l
i
s
t
i
n
g
p
r
i
m
a
r
y
s
t
u
d
i
e
s
(
4
4
)
E
n
d
N
Y
Fig
u
r
e
3
.
Flo
w
d
ia
g
r
am
o
f
th
e
p
r
o
p
o
s
ed
s
ea
r
ch
s
tu
d
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
4
9
-
1
9
6
0
1954
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Ans
wer
i
ng
t
he
re
s
ea
rc
h que
s
t
io
ns
T
h
is
s
tu
d
y
an
aly
s
es p
r
ev
io
u
s
r
esu
lts
to
ad
d
r
ess
th
e
th
r
ee
r
ese
ar
ch
in
q
u
ir
ies alr
ea
d
y
o
u
tlin
ed
.
Q1
: Wh
at
is
th
e
cu
r
r
en
t state
o
f
r
esear
ch
o
n
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
f
o
r
im
b
ala
n
ce
d
d
ata?
Self
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
le
ar
n
in
g
h
as
ac
q
u
ir
ed
s
ig
n
if
ican
t
atten
tio
n
i
n
th
e
r
esear
ch
co
m
m
u
n
ity
f
o
r
ef
f
icien
tly
lear
n
in
g
r
e
p
r
esen
ta
tio
n
s
f
r
o
m
u
n
lab
eled
d
ata.
T
u
et
a
l.
[
6
0
]
in
tr
o
d
u
ce
d
AAG,
wh
ich
m
ak
es
u
s
e
o
f
an
au
x
iliar
y
a
u
g
m
e
n
tatio
n
s
tr
ateg
y
an
d
GNT
-
Xe
n
t
lo
s
s
.
Similar
ly
,
T
ian
et
a
l.
[
6
1
]
p
r
o
p
o
s
ed
co
n
s
tr
ain
e
d
co
n
tr
asti
v
e
d
is
tr
ib
u
tio
n
lear
n
i
n
g
f
o
r
a
n
o
m
aly
d
etec
tio
n
(
C
C
D)
,
an
ap
p
r
o
ac
h
f
o
r
an
o
m
aly
d
etec
tio
n
in
m
ed
ical
p
h
o
to
s
u
s
in
g
lim
ited
co
n
tr
a
s
tiv
e
d
is
tr
ib
u
tio
n
lear
n
in
g
,
wh
ich
f
o
cu
s
es
o
n
lear
n
in
g
f
in
e
-
g
r
ain
e
d
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
th
r
o
u
g
h
co
n
tr
asti
v
e
lear
n
in
g
with
p
r
etex
t
co
n
s
tr
ain
ts
.
Gao
et
a
l.
[
6
2
]
p
r
o
p
o
s
ed
d
is
till
ed
co
n
tr
asti
v
e
lear
n
in
g
(
Dis
C
o
)
as
a
way
to
m
itig
ate
th
e
p
er
f
o
r
m
a
n
ce
d
eter
io
r
atio
n
o
f
SS
L
o
n
lig
h
tweig
h
t
m
o
d
els.
Yao
et
a
l.
[
6
3
]
i
n
v
esti
g
ated
SS
L
o
n
elec
tr
o
n
i
c
h
ea
lth
r
ec
o
r
d
s
u
s
in
g
g
r
a
p
h
k
er
n
el
in
f
o
m
ax
,
s
h
o
wca
s
in
g
th
e
s
u
cc
ess
o
f
co
n
tr
asti
v
e
lear
n
in
g
in
th
is
d
o
m
ai
n
.
Fu
r
th
er
m
o
r
e,
Kan
g
et
a
l.
[
6
4
]
f
o
cu
s
ed
o
n
lo
n
g
-
tailed
lear
n
in
g
a
n
d
im
p
r
o
v
in
g
f
ea
tu
r
e
e
x
tr
ac
to
r
s
an
d
class
if
ier
s
f
o
r
im
b
alan
ce
d
d
ata
t
h
r
o
u
g
h
co
n
tr
asti
v
e
p
r
etr
ain
in
g
an
d
f
ea
tu
r
e
n
o
r
m
a
lizatio
n
.
T
r
äu
b
le
et
a
l.
[
6
5
]
i
n
tr
o
d
u
ce
d
a
n
o
v
el
co
n
t
r
asti
v
e
lo
s
s
f
o
r
b
r
ain
a
g
e
p
r
ed
ictio
n
o
n
3
D
s
tiff
n
ess
m
ap
s
,
aim
in
g
to
im
p
r
o
v
e
g
e
n
er
a
lizatio
n
ac
r
o
s
s
n
o
n
-
u
n
if
o
r
m
ly
d
is
tr
ib
u
ted
d
ata
in
m
ed
ical
im
ag
in
g
d
ata.
Q2
:
W
h
at
is
th
e
m
o
s
t
ef
f
ec
tiv
e
ap
p
r
o
ac
h
u
s
in
g
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
to
ad
d
r
ess
im
b
alan
ce
d
d
ata
class
if
icatio
n
?
B
ased
o
n
c
o
n
tr
asti
v
e
SS
L
,
r
esear
ch
er
s
h
a
v
e
p
r
o
p
o
s
ed
s
ev
er
a
l
in
n
o
v
ativ
e
tec
h
n
iq
u
es
an
d
s
tr
ateg
ies
to
en
h
an
ce
r
ep
r
esen
tatio
n
lear
n
i
n
g
.
T
h
ese
ap
p
r
o
ac
h
es
aim
to
im
p
r
o
v
e
th
e
r
o
b
u
s
tn
ess
o
f
cla
s
s
if
ier
s
,
p
ar
ticu
lar
ly
in
th
e
co
n
tex
t o
f
s
k
ewe
d
ca
te
g
o
r
y
d
is
tr
ib
u
tio
n
s
.
B
y
ad
d
r
ess
in
g
th
ese
ch
allen
g
es,
th
e
y
ef
f
e
ctiv
ely
co
n
tr
ib
u
te
to
s
o
lv
in
g
im
b
alan
ce
d
class
if
icatio
n
is
s
u
es.
Var
io
u
s
m
eth
o
d
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
to
en
h
a
n
ce
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
,
s
u
ch
as d
ev
is
in
g
s
am
p
lin
g
s
tr
ateg
ies
th
at
en
s
u
r
e
m
in
o
r
it
y
class
es
ar
e
ad
e
q
u
ately
r
ep
r
esen
ted
in
t
h
e
co
n
tr
asti
v
e
lea
r
n
in
g
p
r
o
ce
s
s
.
Fo
r
ex
am
p
le,
m
o
d
el
-
Awa
r
e
K
-
ce
n
ter
[
6
6
]
im
p
r
o
v
ed
co
n
tr
asti
v
e
lear
n
in
g
o
n
im
b
alan
ce
d
s
ee
d
d
ata
is
al
s
o
ex
p
lo
r
ed
th
r
o
u
g
h
an
o
p
e
n
-
wo
r
l
d
s
am
p
li
n
g
f
r
am
ewo
r
k
,
wh
ich
s
tr
ateg
ically
s
elec
ts
u
n
lab
eled
d
ata
f
r
o
m
ex
ter
n
al
s
o
u
r
ce
s
to
lear
n
g
en
er
aliza
b
le
,
b
ala
n
c
ed
,
an
d
d
iv
e
r
s
e
r
ep
r
esen
tatio
n
s
.
Yan
g
et
a
l.
[
6
7
]
s
u
g
g
ested
a
n
o
v
el
h
y
p
e
r
g
r
a
p
h
co
n
tr
asti
v
e
lear
n
in
g
m
o
d
el
(
I
S
-
HGCL)
th
at
u
tili
ze
s
h
y
p
er
g
r
ap
h
s
to
tack
le
th
e
p
r
o
b
le
m
s
o
f
im
b
alan
ce
an
d
lo
n
g
-
tail d
is
tr
ib
u
tio
n
in
g
r
ad
u
a
te
d
ev
elo
p
m
e
n
t p
r
e
d
ictio
n
s
.
Ad
ju
s
tin
g
th
e
m
ar
g
in
o
r
we
ig
h
tin
g
th
e
co
n
tr
asti
v
e
lo
s
s
b
ased
o
n
class
d
is
tr
ib
u
tio
n
o
r
s
am
p
le
h
ar
d
n
ess
,
m
ak
es
th
e
m
o
d
el
s
en
s
itiv
e
to
th
e
lear
n
in
g
d
if
f
icu
lty
o
f
d
i
f
f
er
en
t
class
es.
T
h
e
SC
o
R
e
[
6
8
]
f
r
am
ewo
r
k
in
tr
o
d
u
ce
s
s
u
b
m
o
d
u
lar
co
m
b
in
ato
r
ial
lo
s
s
f
u
n
ct
io
n
s
th
at
ef
f
ec
tiv
ely
ad
d
r
ess
th
e
ch
allen
g
es
p
o
s
ed
b
y
class
im
b
alan
ce
.
E
m
p
ir
ic
al
ev
id
en
ce
d
em
o
n
s
tr
ates
th
at
th
ese
g
o
als
s
u
r
p
ass
th
e
m
o
s
t
ad
v
an
ce
d
m
etr
i
c
lear
n
er
s
n
o
w
av
ailab
le
b
y
as
m
u
ch
as
7
.
6
%
in
im
b
alan
ce
d
class
if
icatio
n
task
s
.
W
an
g
et
a
l.
[
6
9
]
th
e
n
o
v
el
f
o
ca
l
C
L
was
p
r
o
p
o
s
ed
with
s
atellite
im
ag
es,
an
d
Alen
ez
i
et
a
l.
[
7
0
]
in
tr
o
d
u
ce
d
th
e
in
n
o
v
ativ
e
W
-
s
h
ap
ed
C
L
m
o
d
el
u
tili
zin
g
s
k
i
n
lesi
o
n
p
h
o
to
s
as
d
atasets
.
Similar
ly
,
Z
h
an
g
et
a
l.
[
7
1
]
ap
p
lied
co
n
tr
asti
v
e
lear
n
in
g
with
a
weig
h
ted
lo
s
s
f
u
n
ctio
n
to
im
b
alan
ce
d
d
atasets
in
th
e
f
ield
o
f
h
ea
lth
ca
r
e.
Au
d
ib
er
t
et
a
l.
[
7
2
]
in
tr
o
d
u
ce
a
n
ew
m
u
lti
-
lab
el
co
n
tr
asti
v
e
lo
s
s
th
at
ad
ap
ts
th
e
co
n
v
en
tio
n
al
co
n
tr
asti
v
e
lear
n
in
g
f
r
am
ewo
r
k
to
h
an
d
le
d
atasets
with
a
lo
n
g
-
tailed
d
is
tr
ib
u
tio
n
b
etter
.
Oth
er
ef
f
ec
tiv
e
s
tr
ateg
y
cu
r
r
en
tly
b
ein
g
u
s
ed
o
r
ac
tiv
e
ly
r
esear
ch
ed
in
clu
d
e
h
y
b
r
i
d
lear
n
in
g
ap
p
r
o
ac
h
es
an
d
ar
ch
itectu
r
al
i
n
n
o
v
atio
n
s
.
T
ah
er
et
a
l.
[
7
3
]
wer
e
ass
ig
n
ed
th
e
g
o
al
o
f
d
e
v
elo
p
in
g
a
f
r
am
ew
o
r
k
th
at
wo
u
ld
im
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
b
y
in
teg
r
atin
g
co
n
tr
asti
v
e
a
n
d
g
en
e
r
ativ
e
task
s
to
lear
n
b
o
th
g
lo
b
al
an
d
l
o
ca
l
p
r
o
p
er
ties
.
Kallid
r
o
m
itis
et
a
l.
[
7
4
]
in
tr
o
d
u
ce
a
n
o
v
el
f
r
am
e
wo
r
k
co
m
b
in
in
g
c
o
n
tr
asti
v
e
l
ea
r
n
in
g
with
n
e
u
r
al
p
r
o
ce
s
s
es
to
en
h
an
ce
tim
e
s
e
r
ies
f
o
r
ec
asti
n
g
with
o
u
t
r
ely
in
g
o
n
p
r
e
-
d
ef
i
n
ed
d
ata
au
g
m
en
tatio
n
s
,
s
h
o
win
g
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
d
iv
er
s
e
d
atasets
.
Yan
g
et
a
l.
[
7
5
]
p
r
o
p
o
s
ed
p
r
o
to
ty
p
ical
co
n
tr
asti
v
e
lear
n
in
g
(
Pro
C
L
)
,
wh
ich
in
teg
r
ates
co
n
tr
asti
v
e
lear
n
in
g
with
clu
s
ter
in
g
an
d
allo
ca
tes
weig
h
ts
to
n
eg
ativ
e
s
am
p
les
b
ased
o
n
th
e
d
is
tan
ce
to
th
e
p
r
o
to
ty
p
e.
T
ab
le
3
b
r
ief
l
y
o
v
er
v
iews
s
o
m
e
o
f
th
e
m
o
s
t
ef
f
ec
tiv
e
m
eth
o
d
s
o
f
r
esear
ch
o
n
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
a
p
p
licatio
n
s
in
im
b
alan
ce
class
if
icatio
n
.
Q3
: Wh
at
ar
e
th
e
m
o
s
t c
r
itical
g
ap
s
an
d
s
h
o
r
tco
m
in
g
s
in
th
e
r
ev
iewe
d
r
esear
ch
?
Self
-
s
u
p
er
v
is
ed
c
o
n
tr
asti
v
e
le
ar
n
in
g
with
v
a
r
io
u
s
s
tr
ateg
ie
s
s
ig
n
if
ican
tly
e
n
h
an
ce
s
r
ep
r
esen
tatio
n
lear
n
in
g
an
d
ad
d
r
ess
es
th
e
im
b
alan
ce
d
d
ata
class
if
icatio
n
p
r
o
b
lem
.
Ho
wev
e
r
,
th
er
e
a
r
e
s
till
n
o
tab
le
g
ap
s
an
d
d
ef
icien
cies
in
th
ese
a
p
p
r
o
ac
h
es
th
at
n
ee
d
to
b
e
ad
d
r
ess
ed
.
T
h
ese
g
ap
s
ar
e
cr
itical
f
o
r
o
n
g
o
in
g
r
esear
ch
a
n
d
th
e
s
u
cc
ess
f
u
l im
p
lem
en
tatio
n
o
f
p
r
a
ctica
l a
p
p
licatio
n
s
.
First,
m
o
d
els
b
ased
o
n
s
elf
-
s
u
p
er
v
is
ed
c
o
n
tr
asti
v
e
lear
n
in
g
,
esp
ec
ially
th
o
s
e
in
v
o
lv
in
g
lar
g
e
-
s
ca
le
d
ata
au
g
m
en
tatio
n
an
d
co
m
p
lex
s
am
p
lin
g
s
tr
ateg
ies,
ca
n
b
e
co
m
p
u
tatio
n
ally
in
ten
s
iv
e
an
d
r
eq
u
ir
e
m
an
y
h
ar
d
war
e
r
eso
u
r
ce
s
[
7
6
]
.
Seco
n
d
,
c
o
n
tr
asti
v
e
lea
r
n
in
g
m
eth
o
d
s
ar
e
s
u
s
ce
p
tib
le
to
d
ata
q
u
al
ity
.
I
n
ca
s
es
wh
er
e
th
e
d
ata
is
n
o
is
y
o
r
co
n
tain
s
m
an
y
o
u
tlier
s
,
th
e
ef
f
ec
tiv
e
n
ess
o
f
co
n
tr
asti
v
e
lear
n
in
g
m
ay
b
e
r
ed
u
ce
d
,
s
o
d
iv
er
s
e
d
ata
au
g
m
e
n
tatio
n
t
ec
h
n
iq
u
es
ar
e
n
ee
d
ed
[
7
7
]
.
L
astl
y
,
m
o
d
els
tr
ain
ed
u
s
in
g
s
elf
-
s
u
p
e
r
v
is
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
ev
is
itin
g
s
elf
-
s
u
p
ervi
s
ed
co
n
tr
a
s
tive
lea
r
n
in
g
fo
r
imb
a
la
n
ce
d
cla
s
s
ifica
tio
n
(
X
ia
o
lin
g
Ga
o
)
1955
co
n
tr
asti
v
e
lear
n
in
g
m
ay
o
v
e
r
f
it
th
e
ch
ar
ac
ter
is
tics
o
f
th
e
tr
ain
in
g
d
ata,
esp
ec
ially
wh
en
u
s
in
g
lar
g
e
am
o
u
n
ts
o
f
au
g
m
en
tatio
n
o
r
s
p
ec
if
ic
s
a
m
p
lin
g
s
tr
ateg
ies,
wh
ich
m
a
y
h
u
r
t g
e
n
er
aliza
tio
n
[
7
8
]
.
T
ab
le
3
.
Su
m
m
a
r
y
o
f
r
ec
e
n
t liter
atu
r
e
o
n
co
n
tr
asti
v
e
SS
L
in
i
m
b
alan
ce
d
class
if
icatio
n
S
t
r
a
t
e
g
y
R
e
f
e
r
e
n
c
e
s
D
a
t
a
s
e
t
A
d
v
a
n
t
a
g
e
D
i
sad
v
a
n
t
a
g
e
A
u
g
m
e
n
t
a
t
i
o
n
S
t
r
a
t
e
g
y
M
o
C
o
-
C
X
R
R
e
f
.
[
7
9
]
C
h
e
X
p
e
r
t
Pre
-
t
r
a
i
n
i
n
g
i
s
p
a
r
t
i
c
u
l
a
r
l
y
a
d
v
a
n
t
a
g
e
o
u
s w
h
e
n
t
h
e
r
e
i
s
a
sca
r
c
i
t
y
o
f
l
a
b
e
l
l
e
d
t
r
a
i
n
i
n
g
d
a
t
a
T
h
e
p
e
r
f
o
r
ma
n
c
e
g
a
i
n
s
d
i
m
i
n
i
sh
a
s
t
h
e
q
u
a
n
t
i
t
y
o
f
l
a
b
e
l
l
e
d
t
r
a
i
n
i
n
g
d
a
t
a
i
n
c
r
e
a
s
e
s
M
e
t
a
-
Le
a
r
n
i
n
g
M
e
d
A
u
g
R
e
f
.
[
8
0
]
C
h
e
X
p
e
r
t
,
U
t
i
l
i
z
e
p
a
t
i
e
n
t
me
t
a
d
a
t
a
f
o
r
c
h
o
o
s
i
n
g
p
o
si
t
i
v
e
p
a
i
r
s
a
n
d
1
4
.
4
%
h
i
g
h
e
r
i
n
t
h
e
a
v
e
r
a
g
e
A
U
C
t
h
a
n
b
a
s
e
l
i
n
e
sel
e
c
t
i
n
g
h
a
r
d
n
e
g
a
t
i
v
e
p
a
i
r
s
u
s
i
n
g
met
a
d
a
t
a
d
i
d
n
o
t
y
i
e
l
d
i
mp
r
o
v
e
me
n
t
s
o
v
e
r
t
h
e
b
a
se
l
i
n
e
H
y
b
r
i
d
Le
a
r
n
i
n
g
M
I
C
Le
R
e
f
.
[
8
1
]
D
e
r
mat
o
l
o
g
y
,
C
h
e
X
p
e
r
t
Th
e
me
t
h
o
d
s
i
g
n
i
f
i
c
a
n
t
l
y
i
mp
r
o
v
e
s
t
h
e
a
c
c
u
r
a
c
y
o
f
m
e
d
i
c
a
l
i
m
a
g
e
c
l
a
ss
i
f
i
e
r
s
I
t
may
s
t
i
l
l
r
e
q
u
i
r
e
a
su
b
s
t
a
n
t
i
a
l
a
mo
u
n
t
o
f
u
n
l
a
b
e
l
e
d
d
a
t
a
t
o
a
c
h
i
e
v
e
o
p
t
i
m
a
l
p
e
r
f
o
r
ma
n
c
e
S
a
mp
l
i
n
g
S
t
r
a
t
e
g
y
M
A
K
R
e
f
.
[
6
7
]
I
mag
e
N
e
t
-
1
0
0
-
LT
E
f
f
e
c
t
i
v
e
l
y
e
n
h
a
n
c
e
s
t
h
e
q
u
a
l
i
t
y
o
f
l
e
a
r
n
e
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
b
y
st
r
a
t
e
g
i
c
a
l
l
y
sel
e
c
t
i
n
g
u
n
l
a
b
e
l
e
d
d
a
t
a
f
r
o
m e
x
t
e
r
n
a
l
s
o
u
r
c
e
s
c
h
a
l
l
e
n
g
e
s re
l
a
t
e
d
t
o
d
a
t
a
i
m
b
a
l
a
n
c
e
a
n
d
d
i
s
t
r
a
c
t
i
o
n
f
r
o
m
o
u
t
-
of
-
d
i
s
t
r
i
b
u
t
i
o
n
sam
p
l
e
s
A
d
j
u
st
i
n
g
c
o
n
t
r
a
s
t
i
v
e
l
o
ss
A
F
C
L
R
e
f
.
[
8
2
]
F
M
N
I
S
T,
I
S
I
C
2
0
1
8
I
t
o
u
t
p
e
r
f
o
r
m
s
C
L
a
n
d
F
C
L
i
n
t
e
r
ms
o
f
b
o
t
h
w
e
i
g
h
t
e
d
a
n
d
u
n
w
e
i
g
h
t
e
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
i
e
s
I
t
r
e
q
u
i
r
e
s
c
a
r
e
f
u
l
t
u
n
i
n
g
o
f
h
y
p
e
r
p
a
r
a
me
t
e
r
s
r
e
s
u
l
t
i
n
g
i
n
a
c
o
m
p
l
i
c
a
t
e
d
t
r
a
i
n
i
n
g
p
r
o
c
e
s
s
S
a
mp
l
i
n
g
S
t
r
a
t
e
g
y
S
D
C
LR
R
e
f
.
[
8
3
]
C
I
F
A
R
-
10
-
LT,
C
I
F
A
R
-
100
-
L
T,
I
mag
e
N
e
t
-
LT
I
t
f
o
c
u
s
e
s
o
n
d
i
f
f
i
c
u
l
t
-
to
-
l
e
a
r
n
samp
l
e
s
,
w
h
i
c
h
e
n
h
a
n
c
e
s
t
h
e
mo
d
e
l
's
a
b
i
l
i
t
y
t
o
g
e
n
e
r
a
l
i
z
e
b
e
t
t
e
r
a
c
r
o
ss
v
a
r
i
o
u
s
c
l
a
ss
e
s
I
t
may
l
e
a
d
t
o
t
h
e
e
x
c
l
u
s
i
o
n
o
f
s
o
m
e
u
sef
u
l
i
n
f
o
r
ma
t
i
o
n
S
a
mp
l
i
n
g
S
t
r
a
t
e
g
y
B
C
L
R
e
f
.
[
8
4
]
C
I
F
A
R
-
10
-
LT,
C
I
F
A
R
-
100
-
L
T,
I
mag
e
N
e
t
-
L
T,
i
N
a
t
u
r
a
l
i
s
t
2
0
1
8
I
n
c
o
r
p
o
r
a
t
e
s
c
l
a
s
s a
v
e
r
a
g
i
n
g
,
w
h
i
c
h
b
a
l
a
n
c
e
s
t
h
e
g
r
a
d
i
e
n
t
c
o
n
t
r
i
b
u
t
i
o
n
s
f
r
o
m n
e
g
a
t
i
v
e
c
l
a
sses
I
t
may
r
e
q
u
i
r
e
m
o
r
e
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
so
u
r
c
e
s
a
n
d
c
a
r
e
f
u
l
t
u
n
i
n
g
o
f
h
y
p
e
r
p
a
r
a
me
t
e
r
s
t
o
a
c
h
i
e
v
e
o
p
t
i
ma
l
p
e
r
f
o
r
m
a
n
c
e
A
d
j
u
st
i
n
g
c
o
n
t
r
a
s
t
i
v
e
l
o
ss
K
C
L
R
e
f
.
[
8
5
]
I
mag
e
N
e
t
-
L
T,
i
N
a
t
u
r
a
l
i
s
t
2
0
1
8
E
f
f
e
c
t
i
v
e
l
y
b
a
l
a
n
c
e
s t
h
e
n
u
m
b
e
r
o
f
p
o
s
i
t
i
v
e
i
n
s
t
a
n
c
e
s
a
c
r
o
s
s c
l
a
ss
e
s
I
t
may
n
o
t
f
u
l
l
y
l
e
v
e
r
a
g
e
a
l
l
a
v
a
i
l
a
b
l
e
i
n
st
a
n
c
e
s
f
r
o
m
t
h
e
s
a
me
c
l
a
ss
f
o
r
p
o
si
t
i
v
e
p
a
i
r
c
o
n
st
r
u
c
t
i
o
n
S
a
mp
l
i
n
g
S
t
r
a
t
e
g
y
CL
R
e
f
.
[
8
6
]
I
S
I
C
2
0
1
8
,
A
P
TO
S
2
0
1
9
En
h
a
n
c
e
d
t
h
e
c
r
o
ss
-
e
n
t
r
o
p
y
me
t
h
o
d
b
y
e
f
f
e
c
t
i
v
e
l
y
d
i
s
t
i
n
g
u
i
sh
i
n
g
mi
n
o
r
i
t
y
a
n
d
ma
j
o
r
i
t
y
c
l
a
sses
i
n
t
h
e
f
e
a
t
u
r
e
s
p
a
c
e
.
Th
e
r
i
s
k
o
f
o
v
e
r
f
i
t
t
i
n
g
,
p
a
r
t
i
c
u
l
a
r
l
y
w
h
e
n
r
e
s
a
mp
l
i
n
g
i
s a
p
p
l
i
e
d
t
o
mi
n
o
r
i
t
y
c
l
a
sse
s
A
r
c
h
i
t
e
c
t
u
r
a
l
i
n
n
o
v
a
t
i
o
n
H
C
Le
R
e
f
.
[
8
7
]
I
S
I
C
2
0
2
0
，
B
R
A
C
S
,
R
EFU
G
E
I
t
a
c
q
u
i
r
e
s
c
o
n
si
s
t
e
n
t
c
h
a
r
a
c
t
e
r
i
st
i
c
s
o
f
t
h
e
d
o
m
i
n
a
n
t
c
l
a
ss a
n
d
a
v
o
i
d
s
b
e
i
n
g
t
r
a
i
n
e
d
o
n
t
h
e
i
n
f
r
e
q
u
e
n
t
a
n
d
v
a
r
i
e
d
mi
n
o
r
i
t
y
.
I
t
may
l
i
m
i
t
i
t
s a
b
i
l
i
t
y
t
o
g
e
n
e
r
a
l
i
z
e
w
e
l
l
t
o
u
n
s
e
e
n
m
i
n
o
r
i
t
y
c
l
a
ss
i
n
s
t
a
n
c
e
s
A
d
j
u
st
i
n
g
c
o
n
t
r
a
s
t
i
v
e
l
o
ss
TSC
R
e
f
.
[
8
8
]
C
I
F
A
R
-
10
-
LT,
C
I
F
A
R
-
100
-
L
T,
I
mag
e
N
e
t
-
L
T,
i
N
a
t
u
r
a
l
i
s
t
2
0
1
8
En
h
a
n
c
e
s t
h
e
se
p
a
r
a
b
i
l
i
t
y
o
f
mi
n
o
r
i
t
y
c
l
a
sse
s a
n
d
l
e
a
d
s
t
o
b
e
t
t
e
r
g
e
n
e
r
a
l
i
z
a
t
i
o
n
i
n
l
o
n
g
-
t
a
i
l
e
d
r
e
c
o
g
n
i
t
i
o
n
t
a
sk
s
M
a
y
n
o
t
y
i
e
l
d
t
h
e
a
n
a
l
y
t
i
c
a
l
o
p
t
i
ma
l
so
l
u
t
i
o
n
f
o
r
p
o
i
n
t
s
o
n
a
h
y
p
e
r
s
p
h
e
r
e
A
r
c
h
i
t
e
c
t
u
r
a
l
i
n
n
o
v
a
t
i
o
n
S
S
C
L
R
e
f
.
[
8
9
]
S
a
l
i
n
a
s
,
P
a
v
i
a
U
n
i
v
e
r
s
i
t
y
,
a
n
d
B
o
t
sw
a
n
a
S
t
r
e
a
m
l
i
n
e
s
t
h
e
t
r
a
i
n
i
n
g
p
r
o
c
e
d
u
r
e
b
y
o
b
v
i
a
t
i
n
g
t
h
e
n
e
c
e
ss
i
t
y
f
o
r
man
u
a
l
a
n
n
o
t
a
t
i
o
n
Th
e
e
f
f
e
c
t
i
v
e
n
e
ss
o
f
t
h
e
a
c
q
u
i
r
e
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
i
s
h
i
g
h
l
y
d
e
p
e
n
d
e
n
t
o
n
t
h
e
f
o
r
m
u
l
a
t
i
o
n
o
f
t
h
e
p
r
e
l
i
mi
n
a
r
y
t
a
s
k
s
H
y
b
r
i
d
Le
a
r
n
i
n
g
Le
v
e
r
a
g
e
s
C
r
o
ss
-
d
o
m
a
i
n
C
N
N
R
e
f
.
[
9
0
]
H
y
p
e
r
sp
e
c
t
r
a
l
i
ma
g
e
s
Ef
f
e
c
t
i
v
e
l
y
c
a
p
t
u
r
e
s
t
r
a
n
sf
e
r
a
b
l
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
f
r
o
m
l
a
r
g
e
a
m
o
u
n
t
s
o
f
u
n
l
a
b
e
l
e
d
h
y
p
e
r
s
p
e
c
t
r
a
l
i
ma
g
e
s
A
si
g
n
i
f
i
c
a
n
t
c
h
a
l
l
e
n
g
e
a
r
i
ses
i
n
mai
n
t
a
i
n
i
n
g
s
p
e
c
t
r
a
l
h
o
m
o
g
e
n
e
i
t
y
w
h
e
n
u
si
n
g
l
a
r
g
e
r
r
e
g
i
o
n
s
f
o
r
p
se
u
d
o
-
l
a
b
e
l
i
n
g
A
r
c
h
i
t
e
c
t
u
r
a
l
i
n
n
o
v
a
t
i
o
n
I
S
D
R
e
f
.
[
9
1
]
I
mag
e
N
e
t
Ef
f
e
c
t
i
v
e
l
y
u
t
i
l
i
z
e
s
a
so
f
t
s
i
mi
l
a
r
i
t
y
a
p
p
r
o
a
c
h
f
o
r
n
e
g
a
t
i
v
e
i
ma
g
e
s
Th
e
i
t
e
r
a
t
i
v
e
d
i
st
i
l
l
a
t
i
o
n
p
r
o
c
e
d
u
r
e
may
n
e
c
e
s
si
t
a
t
e
a
d
d
i
t
i
o
n
a
l
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
s
o
u
r
c
e
s
.
H
y
b
r
i
d
Le
a
r
n
i
n
g
P
r
o
C
o
R
e
f
.
[
9
2
]
I
S
I
C
2
0
1
8
,
A
P
TO
S
2
0
1
9
A
c
o
m
p
r
e
h
e
n
s
i
v
e
a
p
p
r
o
a
c
h
t
o
a
d
d
r
e
ss
t
h
e
i
ss
u
e
o
f
i
mb
a
l
a
n
c
e
I
t
r
e
l
i
e
s
o
n
a
c
o
m
p
l
e
x
f
r
a
mew
o
r
k
i
n
v
o
l
v
i
n
g
m
u
l
t
i
p
l
e
m
o
d
u
l
e
s
4.
2
.
R
esu
lt
s
d
is
cu
s
s
io
n
T
h
is
s
tu
d
y
r
e
v
iews
th
e
s
tate
-
of
-
th
e
-
a
r
t
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
tech
n
iq
u
es
in
v
o
lv
e
d
in
ad
d
r
ess
in
g
im
b
ala
n
ce
d
class
i
f
icatio
n
.
T
h
e
s
tu
d
y
in
v
esti
g
ates
co
n
tr
asti
v
e
lear
n
i
n
g
with
d
if
f
er
en
t
tr
ai
n
in
g
m
eth
o
d
o
l
o
g
ies
in
d
if
f
er
en
t
d
o
wn
s
tr
ea
m
task
s
,
u
ltima
tely
s
h
o
win
g
th
at
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
in
g
f
o
cu
s
es
o
n
im
p
r
o
v
in
g
m
o
d
e
l
r
o
b
u
s
tn
ess
an
d
g
en
e
r
aliza
tio
n
b
y
e
f
f
ec
tiv
ely
u
tili
zin
g
lar
g
e
q
u
a
n
titi
es
o
f
u
n
lab
eled
d
ata,
wh
ich
is
p
ar
ti
cu
lar
ly
b
en
e
f
icial
in
s
ce
n
ar
io
s
wh
er
e
lab
elled
d
ata
f
o
r
r
ar
e
ev
en
ts
o
r
class
es
is
s
ca
r
ce
.
Mo
r
eo
v
e
r
,
we
ca
n
n
o
t
co
m
b
in
e
o
r
m
a
k
e
s
tatis
tical
c
o
m
p
ar
is
o
n
s
o
f
th
e
im
p
ac
ts
o
f
ea
ch
SSL
tech
n
iq
u
e
o
n
p
e
r
f
o
r
m
an
ce
im
p
r
o
v
em
en
t.
T
h
is
is
b
ec
au
s
e
t
h
e
r
esear
c
h
i
n
clu
d
ed
in
o
u
r
an
aly
s
is
u
tili
ze
s
d
is
tin
ct
d
atasets
,
p
r
o
v
id
es v
a
r
io
u
s
p
e
r
f
o
r
m
an
ce
in
d
icato
r
s
,
an
d
ex
am
in
es d
if
f
e
r
en
t a
im
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
4
9
-
1
9
6
0
1956
Sev
er
al
s
u
g
g
esti
o
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
in
im
b
ala
n
ce
d
class
if
icatio
n
u
s
in
g
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
ar
e
r
ec
o
m
m
en
d
ed
to
f
o
cu
s
m
o
r
e
o
n
th
ese
ch
allen
g
e
s
.
W
e
s
u
g
g
est
u
s
in
g
co
n
tr
asti
v
e
SS
L
p
r
e
-
tr
ain
in
g
in
s
tead
o
f
g
e
n
er
ativ
e
SS
L
p
r
e
-
tr
ain
in
g
f
o
r
th
e
class
if
ic
atio
n
ch
allen
g
e
.
Pre
v
io
u
s
an
aly
s
is
r
ev
ea
ls
th
at
in
teg
r
atin
g
r
e
-
s
am
p
lin
g
s
tr
ateg
ies
with
SS
L
tech
n
iq
u
es
is
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
i
n
s
ce
n
a
r
io
s
o
f
s
ev
er
e
class
im
b
alan
ce
an
d
lo
w
d
ata
av
ailab
ilit
y
.
T
h
e
ch
o
ice
o
f
s
am
p
lin
g
p
r
o
ce
d
u
r
es
ca
n
h
av
e
an
i
m
p
ac
t
o
n
co
n
tr
asti
v
e
SS
L
ap
p
r
o
ac
h
es,
s
u
ch
as
M
o
C
o
an
d
SimCLR,
th
at
r
eq
u
ir
e
a
s
ig
n
if
ican
t
q
u
a
n
tity
o
f
n
eg
ativ
e
ex
am
p
les.
T
h
er
ef
o
r
e,
f
in
d
in
g
a
s
o
lu
tio
n
to
r
ed
u
ce
th
e
d
ep
en
d
en
ce
o
n
s
am
p
le
m
eth
o
d
o
lo
g
ies
r
em
ain
s
an
attr
ac
tiv
e
an
d
u
n
r
eso
lv
ed
is
s
u
e.
Hen
ce
,
ad
d
i
tio
n
al
r
esear
ch
is
r
e
q
u
ir
ed
to
ex
p
lo
r
e
m
eth
o
d
s
f
o
r
g
en
er
atin
g
n
eg
ativ
e
s
am
p
les
an
d
im
p
r
o
v
i
n
g
th
e
in
teg
r
atio
n
o
f
SS
L
with
d
o
wn
s
tr
ea
m
task
s
to
b
o
o
s
t
th
e
ef
f
ec
ti
v
en
ess
o
f
SS
L
ap
p
r
o
ac
h
es
i
n
im
b
alan
ce
d
d
o
m
ain
s
.
An
ad
d
itio
n
al
asp
ec
t
th
at
r
eq
u
ir
es
f
u
r
t
h
er
im
p
r
o
v
em
en
t
is
th
e
m
o
d
if
icatio
n
o
f
th
e
c
o
n
tr
asti
v
e
lo
s
s
f
u
n
ctio
n
,
wh
ich
is
cr
u
cial
f
o
r
en
h
a
n
cin
g
th
e
p
er
f
o
r
m
a
n
ce
.
T
h
e
r
esear
c
h
er
s
h
a
v
e
d
ev
elo
p
ed
co
n
tr
asti
v
e
lo
s
s
f
u
n
ctio
n
s
tailo
r
e
d
f
o
r
s
p
ec
if
ic
u
s
es
in
im
b
alan
ce
d
ar
ea
s
,
s
u
ch
as
m
u
ltim
o
d
al
lear
n
in
g
,
lo
ca
l
r
ep
r
esen
tatio
n
lear
n
in
g
,
an
d
m
u
ltis
ca
le
lear
n
in
g
.
L
astl
y
,
in
teg
r
atio
n
wit
h
o
th
er
SS
L
tech
n
iq
u
es,
lik
e
p
r
etex
t
task
s
o
r
clu
s
ter
in
g
-
b
ased
ap
p
r
o
ac
h
es,
ca
n
h
elp
b
etter
h
a
n
d
le
im
b
al
an
ce
d
d
ata.
Me
an
tim
e,
d
ee
p
e
r
in
teg
r
atio
n
with
d
o
m
ain
-
s
p
ec
if
ic
ap
p
licatio
n
s
an
d
th
e
d
e
v
elo
p
m
e
n
t
o
f
n
ew
b
en
ch
m
ar
k
s
th
at
b
etter
r
ef
lec
t
th
e
c
h
allen
g
es
o
f
im
b
alan
ce
d
d
atasets
in
r
ea
l
-
w
o
r
ld
s
ettin
g
s
.
5.
CO
NCLU
SI
O
N
Self
-
s
u
p
er
v
is
ed
r
ep
r
esen
tatio
n
lear
n
in
g
h
as
g
ar
n
er
ed
co
n
s
id
er
ab
le
in
ter
est
in
r
ec
en
t
y
ea
r
s
b
ec
au
s
e
to
its
ab
ilit
y
to
lear
n
f
r
o
m
u
n
lab
eled
d
ata
ef
f
icien
tly
.
T
h
e
s
tu
d
y
o
f
im
b
alan
ce
d
class
if
icatio
n
p
r
esen
ts
n
u
m
e
r
o
u
s
s
ig
n
if
ican
t
an
d
p
r
ess
in
g
ch
allen
g
es.
Self
-
s
u
p
er
v
is
ed
c
o
n
tr
asti
v
e
lear
n
in
g
is
ev
o
lv
i
n
g
r
a
p
id
ly
,
a
n
d
its
ad
ap
tatio
n
to
im
b
alan
ce
d
d
ata
is
a
p
r
o
m
is
in
g
ar
ea
th
at
b
r
i
d
g
es
th
e
d
is
p
ar
ity
b
etwe
en
u
n
s
u
p
er
v
is
ed
lear
n
in
g
ca
p
ab
ilit
ies an
d
s
u
p
er
v
is
ed
lea
r
n
in
g
'
s
n
ee
d
f
o
r
lab
elled
d
ata.
T
h
is
p
ap
er
c
o
m
p
r
e
h
en
s
iv
ely
r
ev
iews
im
b
alan
ce
class
if
icatio
n
m
eth
o
d
s
b
ased
o
n
s
elf
-
s
u
p
er
v
is
ed
co
n
tr
asti
v
e
lear
n
i
n
g
,
co
v
e
r
i
n
g
th
e
m
o
s
t
p
o
p
u
lar
co
n
tr
asti
v
e
lear
n
in
g
f
r
a
m
ewo
r
k
s
an
d
co
n
s
tr
u
ctio
n
m
ec
h
an
is
m
s
.
I
n
ad
d
itio
n
,
we
p
r
esen
ted
a
co
n
cise
s
u
m
m
ar
y
o
f
th
e
is
s
u
e
o
f
class
im
b
alan
ce
an
d
th
e
latest
ap
p
r
o
ac
h
es
to
ad
d
r
ess
it.
T
h
is
liter
atu
r
e
r
e
v
iew
in
clu
d
es
a
r
ea
s
o
n
ab
le
s
ea
r
ch
m
eth
o
d
,
wit
h
a
lo
w
p
r
o
b
ab
ilit
y
o
f
m
is
s
in
g
ar
ticles
an
d
h
ig
h
s
cien
tific
v
alu
e.
W
e
ca
teg
o
r
ize
d
th
e
SS
L
ap
p
r
o
ac
h
es
an
d
ex
t
r
ac
ted
b
en
e
f
its
an
d
lim
its
f
r
o
m
e
x
is
tin
g
liter
atu
r
e
to
f
o
r
m
u
late
r
ec
o
m
m
en
d
at
io
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
T
h
e
f
o
llo
win
g
s
tu
d
ies
s
h
o
u
ld
in
c
o
r
p
o
r
ate
im
p
r
o
v
e
d
s
am
p
lin
g
an
d
au
g
m
en
tatio
n
te
ch
n
iq
u
es,
as
well
as
a
n
ad
a
p
tiv
e
co
n
t
r
asti
v
e
lo
s
s
f
u
n
ctio
n
,
to
ex
p
ed
ite
th
e
i
d
en
t
if
icatio
n
o
f
o
p
tim
al
m
eth
o
d
s
.
As
r
esear
ch
o
n
h
i
g
h
-
d
im
en
s
io
n
al
im
b
alan
ce
d
d
ata
is
h
ig
h
ly
s
ig
n
if
ican
t,
we
h
o
p
e
to
g
u
id
e
r
esear
ch
e
r
s
in
ter
este
d
in
ex
p
l
o
r
in
g
co
n
tr
asti
v
e
lear
n
in
g
tec
h
n
iq
u
es
to
ex
ten
d
to
m
o
r
e
im
b
alan
ce
f
iel
d
s
,
s
u
ch
as o
b
ject
d
etec
tio
n
,
a
n
d
im
ag
e
s
eg
m
e
n
tatio
n
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
r
esear
ch
was
s
u
p
p
o
r
ted
b
y
th
e
Min
is
tr
y
o
f
Hig
h
er
E
d
u
ca
tio
n
(
MO
HE
)
,
Ma
lay
s
ia
th
r
o
u
g
h
Fu
n
d
am
en
tal
R
esear
ch
Gr
an
t
Sch
em
e
(
FR
GS
/1
/2
0
2
2
/I
C
T
0
2
/UI
T
M/0
2
/8
)
.
Du
e
ac
k
n
o
wled
g
em
en
t
is
also
ac
co
r
d
ed
to
th
e
C
o
lleg
e
o
f
C
o
m
p
u
tin
g
,
I
n
f
o
r
m
atics
an
d
Ma
th
em
atics,
Un
iv
er
s
iti
T
ek
n
o
lo
g
i
MA
R
A
Sh
ah
Alam
,
f
o
r
s
u
p
p
o
r
tin
g
th
is
s
tu
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
S
i
n
g
h
R
a
w
a
t
a
n
d
A
.
K
u
mar
M
i
sh
r
a
,
“
R
e
v
i
e
w
o
f
m
e
t
h
o
d
s
f
o
r
h
a
n
d
l
i
n
g
c
l
a
ss
-
i
mb
a
l
a
n
c
e
d
i
n
c
l
a
ssi
f
i
c
a
t
i
o
n
p
r
o
b
l
e
ms,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
D
a
t
a
,
En
g
i
n
e
e
ri
n
g
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
2
0
2
2
,
p
p
.
3
–
1
4
.
[
2
]
N
.
U
.
N
i
a
z
,
K
.
M
.
N
.
S
h
a
h
a
r
i
a
r
,
a
n
d
M
.
J.
A
.
P
a
t
w
a
r
y
,
“
C
l
a
ss
i
mb
a
l
a
n
c
e
p
r
o
b
l
e
ms
i
n
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
:
a
r
e
v
i
e
w
o
f
m
e
t
h
o
d
s
a
n
d
f
u
t
u
r
e
c
h
a
l
l
e
n
g
e
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
i
n
g
A
d
v
a
n
c
e
m
e
n
t
s
,
M
a
r
.
2
0
2
2
,
p
p
.
4
8
5
–
4
9
0
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
4
2
9
5
4
.
3
5
4
3
0
2
4
.
[
3
]
L.
W
a
n
g
,
M
.
H
a
n
,
X
.
Li
,
N
.
Z
h
a
n
g
,
a
n
d
H
.
C
h
e
n
g
,
“
R
e
v
i
e
w
o
f
c
l
a
ss
i
f
i
c
a
t
i
o
n
me
t
h
o
d
s
o
n
u
n
b
a
l
a
n
c
e
d
d
a
t
a
s
e
t
s,”
I
EEE
Ac
c
e
ss
,
v
o
l
.
9
,
p
p
.
6
4
6
0
6
–
6
4
6
2
8
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
0
7
4
2
4
3
.
[
4
]
K
.
R
.
M
.
F
e
r
n
a
n
d
o
a
n
d
C
.
P
.
Ts
o
k
o
s
,
“
D
y
n
a
m
i
c
a
l
l
y
w
e
i
g
h
t
e
d
b
a
l
a
n
c
e
d
l
o
ss
:
c
l
a
ss i
mb
a
l
a
n
c
e
d
l
e
a
r
n
i
n
g
a
n
d
c
o
n
f
i
d
e
n
c
e
c
a
l
i
b
r
a
t
i
o
n
o
f
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
I
E
EE
T
r
a
n
sa
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
a
n
d
L
e
a
r
n
i
n
g
S
y
st
e
m
s
,
v
o
l
.
3
3
,
n
o
.
7
,
p
p
.
2
9
4
0
–
2
9
5
1
,
Ju
l
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS.
2
0
2
0
.
3
0
4
7
3
3
5
.
[
5
]
K
.
O
h
r
i
a
n
d
M
.
K
u
mar
,
“
R
e
v
i
e
w
o
n
sel
f
-
su
p
e
r
v
i
s
e
d
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
K
n
o
w
l
e
d
g
e
-
Ba
s
e
d
S
y
s
t
e
m
s
,
v
o
l
.
2
2
4
,
J
u
l
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
n
o
sy
s.
2
0
2
1
.
1
0
7
0
9
0
.
[
6
]
Z.
Z
h
a
o
,
L
.
A
l
z
u
b
a
i
d
i
,
J.
Zh
a
n
g
,
Y
.
D
u
a
n
,
a
n
d
Y
.
G
u
,
“
A
c
o
m
p
a
r
i
so
n
r
e
v
i
e
w
o
f
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
n
d
sel
f
-
s
u
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
:
d
e
f
i
n
i
t
i
o
n
s,
a
p
p
l
i
c
a
t
i
o
n
s,
a
d
v
a
n
t
a
g
e
s
a
n
d
l
i
m
i
t
a
t
i
o
n
s,
”
E
x
p
e
rt
S
y
s
t
e
m
s
w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
4
2
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
3
.
1
2
2
8
0
7
.
[
7
]
S
.
G
a
o
e
t
a
l
.
,
“
L
e
a
r
n
i
n
g
i
n
t
h
e
w
i
l
d
:
t
o
w
a
r
d
s
l
e
v
e
r
a
g
i
n
g
u
n
l
a
b
e
l
e
d
d
a
t
a
f
o
r
e
f
f
e
c
t
i
v
e
l
y
t
u
n
i
n
g
p
r
e
-
t
r
a
i
n
e
d
c
o
d
e
m
o
d
e
l
s
,
”
i
n
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
E
E
E
/
A
C
M
4
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
o
f
t
w
a
r
e
E
n
g
i
n
e
e
r
i
n
g
,
A
p
r
.
2
0
2
4
,
p
p
.
1
–
13
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
9
7
5
0
3
.
3
6
3
9
2
1
6
.
[
8
]
X
.
Li
u
e
t
a
l
.
,
“
S
e
l
f
-
su
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
:
g
e
n
e
r
a
t
i
v
e
o
r
c
o
n
t
r
a
s
t
i
v
e
,
”
I
E
EE
T
r
a
n
s
a
c
t
i
o
n
s o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
E
n
g
i
n
e
e
r
i
n
g
,
p
.
1
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
R
ev
is
itin
g
s
elf
-
s
u
p
ervi
s
ed
co
n
tr
a
s
tive
lea
r
n
in
g
fo
r
imb
a
la
n
ce
d
cla
s
s
ifica
tio
n
(
X
ia
o
lin
g
Ga
o
)
1957
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
2
1
.
3
0
9
0
8
6
6
.
[
9
]
T.
C
h
e
n
,
S
.
K
o
r
n
b
l
i
t
h
,
M
.
N
o
r
o
u
z
i
,
a
n
d
G
.
H
i
n
t
o
n
,
“
A
s
i
m
p
l
e
f
r
a
mew
o
r
k
f
o
r
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
o
f
v
i
s
u
a
l
r
e
p
r
e
sen
t
a
t
i
o
n
s,”
i
n
3
7
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
I
C
ML
2
0
2
0
,
2
0
2
0
,
p
p
.
1
5
7
5
–
1
5
8
5
.
[
1
0
]
J.
G
u
i
e
t
a
l
.
,
“
A
s
u
r
v
e
y
o
n
se
l
f
-
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
:
a
l
g
o
r
i
t
h
ms
,
a
p
p
l
i
c
a
t
i
o
n
s,
a
n
d
f
u
t
u
r
e
t
r
e
n
d
s
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
2
4
.
3
4
1
5
1
1
2
.
[
1
1
]
J.
Li
,
P
.
Z
h
o
u
,
C
.
X
i
o
n
g
,
a
n
d
S
.
C
.
H
.
H
o
i
,
“
P
r
o
t
o
t
y
p
i
c
a
l
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
o
f
u
n
su
p
e
r
v
i
s
e
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s,
”
i
n
I
C
L
R
2
0
2
1
-
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
L
e
a
r
n
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s
,
2
0
2
1
,
p
p
.
1
–
1
6
.
[
1
2
]
C
.
D
u
,
Y
.
W
a
n
g
,
C
.
W
a
n
g
,
C
.
S
h
i
,
a
n
d
B
.
X
i
a
o
,
“
S
e
l
e
c
t
i
v
e
f
e
a
t
u
r
e
c
o
n
n
e
c
t
i
o
n
m
e
c
h
a
n
i
sm
:
C
o
n
c
a
t
e
n
a
t
i
n
g
mu
l
t
i
-
l
a
y
e
r
C
N
N
f
e
a
t
u
r
e
s w
i
t
h
a
f
e
a
t
u
r
e
s
e
l
e
c
t
o
r
,
”
P
a
t
t
e
rn
R
e
c
o
g
n
i
t
i
o
n
L
e
t
t
e
rs
,
v
o
l
.
1
2
9
,
p
p
.
1
0
8
–
1
1
4
,
Ja
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
r
e
c
.
2
0
1
9
.
1
1
.
0
1
5
.
[
1
3
]
K
.
H
e
,
H
.
F
a
n
,
Y
.
W
u
,
S
.
X
i
e
,
a
n
d
R
.
G
i
r
sh
i
c
k
,
“
M
o
me
n
t
u
m
c
o
n
t
r
a
s
t
f
o
r
u
n
s
u
p
e
r
v
i
se
d
v
i
s
u
a
l
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EE
E
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
2
0
2
0
,
p
p
.
9
7
2
6
–
9
7
3
5
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
2
6
0
0
.
2
0
2
0
.
0
0
9
7
5
.
[
1
4
]
L.
Ji
n
g
a
n
d
Y
.
T
i
a
n
,
“
S
e
l
f
-
s
u
p
e
r
v
i
se
d
v
i
s
u
a
l
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
w
i
t
h
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s:
a
su
r
v
e
y
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s o
n
P
a
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
3
,
n
o
.
1
1
,
p
p
.
4
0
3
7
–
4
0
5
8
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
2
0
.
2
9
9
2
3
9
3
.
[
1
5
]
S
.
S
r
i
n
i
v
a
s,
R
.
K
.
S
a
r
v
a
d
e
v
a
b
h
a
t
l
a
,
K
.
R
.
M
o
p
u
r
i
,
N
.
P
r
a
b
h
u
,
S
.
S
.
S
.
K
r
u
t
h
i
v
e
n
t
i
,
a
n
d
R
.
V
.
B
a
b
u
,
“
A
n
i
n
t
r
o
d
u
c
t
i
o
n
t
o
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
s
f
o
r
c
o
m
p
u
t
e
r
v
i
si
o
n
,
”
D
e
e
p
L
e
a
r
n
i
n
g
f
o
r
Me
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
si
s
,
p
p
.
2
5
–
5
2
,
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
B
9
7
8
-
0
-
12
-
8
1
0
4
0
8
-
8
.
0
0
0
0
3
-
1.
[
1
6
]
S.
-
C
.
H
u
a
n
g
,
A
.
P
a
r
e
e
k
,
M
.
Je
n
se
n
,
M
.
P
.
L
u
n
g
r
e
n
,
S
.
Y
e
u
n
g
,
a
n
d
A
.
S
.
C
h
a
u
d
h
a
r
i
,
“
S
e
l
f
-
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
f
o
r
me
d
i
c
a
l
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
:
a
sy
s
t
e
m
a
t
i
c
r
e
v
i
e
w
a
n
d
i
mp
l
e
m
e
n
t
a
t
i
o
n
g
u
i
d
e
l
i
n
e
s,”
n
p
j
D
i
g
i
t
a
l
M
e
d
i
c
i
n
e
,
v
o
l
.
6
,
n
o
.
1
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
7
4
6
-
0
2
3
-
0
0
8
1
1
-
0.
[
1
7
]
A
.
A
g
i
z
a
,
M
.
N
e
se
e
m,
a
n
d
S
.
R
e
d
a
,
“
M
TL
o
R
A
:
a
l
o
w
-
r
a
n
k
a
d
a
p
t
a
t
i
o
n
a
p
p
r
o
a
c
h
f
o
r
e
f
f
i
c
i
e
n
t
m
u
l
t
i
-
t
a
s
k
l
e
a
r
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
2
0
2
4
,
p
p
.
1
6
1
9
6
–
1
6
2
0
5
.
[
1
8
]
D
.
W
o
l
f
e
t
a
l
.
,
“
S
e
l
f
-
su
p
e
r
v
i
s
e
d
p
r
e
-
t
r
a
i
n
i
n
g
w
i
t
h
c
o
n
t
r
a
st
i
v
e
a
n
d
mas
k
e
d
a
u
t
o
e
n
c
o
d
e
r
me
t
h
o
d
s
f
o
r
d
e
a
l
i
n
g
w
i
t
h
sma
l
l
d
a
t
a
se
t
s
i
n
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
m
e
d
i
c
a
l
i
ma
g
i
n
g
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
rt
s
,
v
o
l
.
1
3
,
n
o
.
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
8
-
0
2
3
-
4
6
4
3
3
-
0.
[
1
9
]
V
.
R
a
n
i
,
S
.
T.
N
a
b
i
,
M
.
K
u
m
a
r
,
A
.
M
i
t
t
a
l
,
a
n
d
K
.
K
u
m
a
r
,
“
S
e
l
f
-
su
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
:
a
s
u
c
c
i
n
c
t
r
e
v
i
e
w
,
”
Ar
c
h
i
v
e
s
o
f
C
o
m
p
u
t
a
t
i
o
n
a
l
Me
t
h
o
d
s i
n
En
g
i
n
e
e
r
i
n
g
,
v
o
l
.
3
0
,
n
o
.
4
,
p
p
.
2
7
6
1
–
2
7
7
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
8
3
1
-
0
2
3
-
0
9
8
8
4
-
2.
[
2
0
]
X
.
C
h
e
n
,
H
.
F
a
n
,
R
.
G
i
r
sh
i
c
k
,
a
n
d
K
.
H
e
,
“
I
mp
r
o
v
e
d
b
a
s
e
l
i
n
e
s
w
i
t
h
m
o
me
n
t
u
m
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
2
0
0
3
.
0
4
2
9
7
,
2
0
2
0
.
[
2
1
]
M
.
C
a
r
o
n
,
I
.
M
i
sr
a
,
J
.
M
a
i
r
a
l
,
P
.
G
o
y
a
l
,
P
.
B
o
j
a
n
o
w
sk
i
,
a
n
d
A
.
J
o
u
l
i
n
,
“
U
n
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
o
f
v
i
s
u
a
l
f
e
a
t
u
r
e
s
b
y
c
o
n
t
r
a
s
t
i
ng
c
l
u
st
e
r
a
ss
i
g
n
me
n
t
s,”
A
d
v
a
n
c
e
s
i
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
v
o
l
.
2
0
2
0
-
D
e
c
e
m,
2
0
2
0
.
[
2
2
]
Y
.
Ti
a
n
,
D
.
K
r
i
s
h
n
a
n
,
a
n
d
P
.
I
so
l
a
,
“
C
o
n
t
r
a
s
t
i
v
e
m
u
l
t
i
v
i
e
w
c
o
d
i
n
g
,
”
i
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
–
EC
C
V
2
0
2
0
:
1
6
t
h
Eu
r
o
p
e
a
n
C
o
n
f
e
re
n
c
e
,
G
l
a
s
g
o
w
,
U
K
,
A
u
g
u
st
2
3
–
2
8
,
2
0
2
0
,
P
ro
c
e
e
d
i
n
g
s
,
P
a
r
t
XI
1
6
,
2
0
2
0
,
p
p
.
7
7
6
–
794
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
5
8
6
2
1
-
8
_
4
5
.
[
2
3
]
Y
.
O
u
y
a
n
g
,
T.
F
e
n
g
,
R
.
G
a
o
,
Y
.
X
u
,
a
n
d
J.
L
i
u
,
“
P
r
e
d
i
c
t
i
o
n
o
f
g
r
a
d
u
a
t
i
o
n
d
e
v
e
l
o
p
me
n
t
b
a
se
d
o
n
h
y
p
e
r
g
r
a
p
h
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
w
i
t
h
i
m
b
a
l
a
n
c
e
d
sam
p
l
i
n
g
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
8
9
8
8
1
–
8
9
8
9
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
3
.
3
3
0
1
8
7
8
.
[
2
4
]
L.
L
o
g
e
sw
a
r
a
n
a
n
d
H
.
L
e
e
,
“
A
n
e
f
f
i
c
i
e
n
t
f
r
a
m
e
w
o
r
k
f
o
r
l
e
a
r
n
i
n
g
se
n
t
e
n
c
e
r
e
p
r
e
se
n
t
a
t
i
o
n
s
,
”
i
n
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
L
e
a
r
n
i
n
g
R
e
p
r
e
se
n
t
a
t
i
o
n
s
,
I
C
L
R
2
0
1
8
-
C
o
n
f
e
r
e
n
c
e
T
r
a
c
k
Pro
c
e
e
d
i
n
g
s
,
2
0
1
8
,
p
p
.
1
–
1
6
.
[
2
5
]
C
.
Y
.
C
h
u
a
n
g
,
J.
R
o
b
i
n
so
n
,
L
.
Y
e
n
-
C
h
e
n
,
A
.
T
o
r
r
a
l
b
a
,
a
n
d
S
.
J
e
g
e
l
k
a
,
“
D
e
b
i
a
s
e
d
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
,
”
A
d
v
a
n
c
e
s
i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
v
o
l
.
2
0
2
0
-
D
e
c
e
m,
2
0
2
0
.
[
2
6
]
A
.
B
a
e
v
s
k
i
,
H
.
Zh
o
u
,
A
.
M
o
h
a
me
d
,
a
n
d
M
.
A
u
l
i
,
“
w
a
v
2
v
e
c
2
.
0
:
A
f
r
a
mew
o
r
k
f
o
r
s
e
l
f
-
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
o
f
s
p
e
e
c
h
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
,
”
Ad
v
a
n
c
e
s
i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
v
o
l
.
2
0
2
0
-
D
e
c
e
m,
2
0
2
0
.
[
2
7
]
L.
W
a
n
g
a
n
d
A
.
v
a
n
d
e
n
O
o
r
d
,
“
M
u
l
t
i
-
f
o
r
mat
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
o
f
a
u
d
i
o
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s,”
p
r
e
p
ri
n
t
a
rX
i
v
:
2
1
0
3
.
0
6
5
0
8
,
2
0
2
1
.
[
2
8
]
S
.
M
a
,
Z.
Ze
n
g
,
D
.
M
c
D
u
f
f
,
a
n
d
Y
.
S
o
n
g
,
“
A
c
t
i
v
e
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
o
f
a
u
d
i
o
-
v
i
s
u
a
l
v
i
d
e
o
r
e
p
r
e
se
n
t
a
t
i
o
n
s,
”
i
n
I
C
L
R
2
0
2
1
-
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
L
e
a
r
n
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s
,
2
0
2
1
,
p
p
.
1
–
1
9
.
[
2
9
]
P
.
M
o
r
g
a
d
o
,
N
.
V
a
s
c
o
n
c
e
l
o
s,
a
n
d
I
.
M
i
sr
a
,
“
A
u
d
i
o
-
v
i
s
u
a
l
i
n
st
a
n
c
e
d
i
s
c
r
i
m
i
n
a
t
i
o
n
w
i
t
h
c
r
o
ss
-
m
o
d
a
l
a
g
r
e
e
me
n
t
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
Pa
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
2
0
2
1
,
p
p
.
1
2
4
7
0
–
1
2
4
8
1
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
6
4
3
7
.
2
0
2
1
.
0
1
2
2
9
.
[
3
0
]
J.
B
.
G
r
i
l
l
e
t
a
l
.
,
“
B
o
o
t
s
t
r
a
p
y
o
u
r
o
w
n
l
a
t
e
n
t
a
n
e
w
a
p
p
r
o
a
c
h
t
o
se
l
f
-
su
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
,
”
Ad
v
a
n
c
e
s
i
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ssi
n
g
S
y
s
t
e
m
s
,
v
o
l
.
2
0
2
0
-
D
e
c
e
m
,
2
0
2
0
.
[
3
1
]
A
.
S
h
r
i
v
a
st
a
v
a
,
A
.
G
u
p
t
a
,
a
n
d
R
.
G
i
r
sh
i
c
k
,
“
D
i
me
n
si
o
n
a
l
i
t
y
r
e
d
u
c
t
i
o
n
b
y
l
e
a
r
n
i
n
g
a
n
i
n
v
a
r
i
a
n
t
m
a
p
p
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
2
0
1
6
,
v
o
l
.
2
0
1
6
-
D
e
c
e
m,
p
p
.
7
6
1
–
7
6
9
.
[
3
2
]
X
.
W
a
n
g
,
R
.
Z
h
a
n
g
,
C
.
S
h
e
n
,
T.
K
o
n
g
,
a
n
d
L.
L
i
,
“
D
e
n
se
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
f
o
r
se
l
f
-
su
p
e
r
v
i
s
e
d
v
i
su
a
l
p
r
e
-
t
r
a
i
n
i
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EE
E
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
2
0
2
1
,
p
p
.
3
0
2
3
–
3
0
3
2
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
6
4
3
7
.
2
0
2
1
.
0
0
3
0
4
.
[
3
3
]
K
.
K
o
t
a
r
,
G
.
I
l
h
a
r
c
o
,
L.
S
c
h
mi
d
t
,
K
.
Eh
s
a
n
i
,
a
n
d
R
.
M
o
t
t
a
g
h
i
,
“
C
o
n
t
r
a
st
i
n
g
c
o
n
t
r
a
s
t
i
v
e
sel
f
-
s
u
p
e
r
v
i
se
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
p
i
p
e
l
i
n
e
s
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE/
C
VF
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
2
0
2
2
,
p
p
.
9
9
2
9
–
9
9
3
9
,
d
o
i
:
1
0
.
1
1
0
9
/
i
c
c
v
4
8
9
2
2
.
2
0
2
1
.
0
0
9
8
0
.
[
3
4
]
N
.
V
C
h
a
w
l
a
,
K
.
W
.
B
o
w
y
e
r
,
L
.
O
.
H
a
l
l
,
a
n
d
W
.
P
.
K
e
g
e
l
me
y
e
r
,
“
S
M
O
T
E:
sy
n
t
h
e
t
i
c
mi
n
o
r
i
t
y
o
v
e
r
-
samp
l
i
n
g
t
e
c
h
n
i
q
u
e
,
”
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
s
e
a
rc
h
,
v
o
l
.
1
6
,
p
p
.
3
2
1
–
3
5
7
,
2
0
0
2
,
d
o
i
:
1
0
.
1
6
1
3
/
j
a
i
r
.
9
5
3
.
[
3
5
]
T.
W
a
n
g
a
n
d
P
.
I
so
l
a
,
“
U
n
d
e
r
st
a
n
d
i
n
g
c
o
n
t
r
a
s
t
i
v
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
t
h
r
o
u
g
h
a
l
i
g
n
me
n
t
a
n
d
u
n
i
f
o
r
m
i
t
y
o
n
t
h
e
h
y
p
e
r
s
p
h
e
r
e
,
”
i
n
3
7
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
I
C
ML
2
0
2
0
,
2
0
2
0
,
v
o
l
.
P
a
r
t
F
1
6
8
1
4
,
p
p
.
9
8
7
1
–
9
8
8
1
.
[
3
6
]
H
.
H
a
n
,
W
.
Y
.
W
a
n
g
,
a
n
d
B
.
H
.
M
a
o
,
“
B
o
r
d
e
r
l
i
n
e
-
S
M
O
T
E:
a
n
e
w
o
v
e
r
-
samp
l
i
n
g
m
e
t
h
o
d
i
n
i
m
b
a
l
a
n
c
e
d
d
a
t
a
se
t
s
l
e
a
r
n
i
n
g
,
”
L
e
c
t
u
re
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
6
4
4
,
n
o
.
P
A
R
T
I
,
p
p
.
8
7
8
–
8
8
7
,
2
0
0
5
,
d
o
i
:
1
0
.
1
0
0
7
/
1
1
5
3
8
0
5
9
_
9
1
.
[
3
7
]
H
.
H
e
,
Y
.
B
a
i
,
E
.
A
.
G
a
r
c
i
a
,
a
n
d
S
.
L
i
,
“
A
D
A
S
Y
N
:
A
d
a
p
t
i
v
e
s
y
n
t
h
e
t
i
c
s
a
m
p
l
i
n
g
a
p
p
r
o
a
c
h
f
o
r
i
m
b
a
l
a
n
c
e
d
l
e
a
r
n
i
n
g
,
”
i
n
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
r
e
n
c
e
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
,
2
0
0
8
,
p
p
.
1
3
2
2
–
1
3
2
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
J
C
N
N
.
2
0
0
8
.
4
6
3
3
9
6
9
.
[
3
8
]
C
.
B
u
n
k
h
u
m
p
o
r
n
p
a
t
,
K
.
S
i
n
a
p
i
r
o
ms
a
r
a
n
,
a
n
d
C
.
Lu
r
s
i
n
s
a
p
,
“
S
a
f
e
-
l
e
v
e
l
-
S
M
O
TE
:
safe
-
l
e
v
e
l
-
s
y
n
t
h
e
t
i
c
mi
n
o
r
i
t
y
o
v
e
r
-
sa
mp
l
i
n
g
t
e
c
h
n
i
q
u
e
f
o
r
h
a
n
d
l
i
n
g
t
h
e
c
l
a
ss
i
m
b
a
l
a
n
c
e
d
p
r
o
b
l
e
m,
”
L
e
c
t
u
re
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
s
u
b
s
e
ri
e
s
L
e
c
t
u
re
N
o
t
e
s
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
re
N
o
t
e
s
i
n
Bi
o
i
n
f
o
rm
a
t
i
c
s)
,
v
o
l
.
5
4
7
6
LN
A
I
,
p
p
.
4
7
5
–
4
8
2
,
2
0
0
9
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
6
4
2
-
0
1
3
0
7
-
2
_
4
3
.
[
3
9
]
W
.
A
.
R
i
v
e
r
a
,
“
N
o
i
se
r
e
d
u
c
t
i
o
n
a
p
r
i
o
r
i
s
y
n
t
h
e
t
i
c
o
v
e
r
-
sam
p
l
i
n
g
f
o
r
c
l
a
ss
i
mb
a
l
a
n
c
e
d
d
a
t
a
set
s
,
”
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
4
0
8
,
p
p
.
1
4
6
–
1
6
1
,
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
s.
2
0
1
7
.
0
4
.
0
4
6
.
[
4
0
]
S
.
S
h
a
r
ma
,
C
.
B
e
l
l
i
n
g
e
r
,
B
.
K
r
a
w
c
z
y
k
,
O
.
Za
i
a
n
e
,
a
n
d
N
.
J
a
p
k
o
w
i
c
z
,
“
S
y
n
t
h
e
t
i
c
o
v
e
r
sam
p
l
i
n
g
w
i
t
h
t
h
e
m
a
j
o
r
i
t
y
c
l
a
ss:
a
n
e
w
p
e
r
s
p
e
c
t
i
v
e
o
n
h
a
n
d
l
i
n
g
e
x
t
r
e
me
i
m
b
a
l
a
n
c
e
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
-
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
D
a
t
a
Mi
n
i
n
g
,
I
C
D
M
,
2
0
1
8
,
v
o
l
.
2
0
1
8
-
N
o
v
e
m,
p
p
.
4
4
7
–
4
5
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
D
M
.
2
0
1
8
.
0
0
0
6
0
.
[
4
1
]
B
.
K
r
a
w
c
z
y
k
,
M
.
K
o
z
i
a
r
s
k
i
,
a
n
d
M
.
W
o
z
n
i
a
k
,
“
R
a
d
i
a
l
-
b
a
s
e
d
o
v
e
r
s
a
m
p
l
i
n
g
f
o
r
m
u
l
t
i
c
l
a
s
s
i
m
b
a
l
a
n
c
e
d
d
a
t
a
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
a
n
d
L
e
a
r
n
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
1
,
n
o
.
8
,
p
p
.
2
8
1
8
–
2
8
3
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
T
N
N
L
S
.
2
0
1
9
.
2
9
1
3
6
7
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
4
9
-
1
9
6
0
1958
[
4
2
]
M
.
K
.
B
.
K
o
z
i
a
r
s
k
i
M
i
c
h
a
ł
a
n
d
W
o
ź
n
i
a
k
,
“
C
o
m
b
i
n
e
d
c
l
e
a
n
i
n
g
a
n
d
r
e
sa
mp
l
i
n
g
a
l
g
o
r
i
t
h
m
f
o
r
m
u
l
t
i
-
c
l
a
ss
i
m
b
a
l
a
n
c
e
d
d
a
t
a
w
i
t
h
l
a
b
e
l
n
o
i
se,
”
K
n
o
w
l
e
d
g
e
-
B
a
se
d
S
y
s
t
e
m
s
,
v
o
l
.
2
0
4
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
n
o
sy
s.
2
0
2
0
.
1
0
6
2
2
3
.
[
4
3
]
L.
A
b
d
i
a
n
d
S
.
H
a
sh
e
mi
,
“
T
o
c
o
mb
a
t
mu
l
t
i
-
c
l
a
ss
i
m
b
a
l
a
n
c
e
d
p
r
o
b
l
e
ms
b
y
m
e
a
n
s
o
f
o
v
e
r
-
sa
mp
l
i
n
g
t
e
c
h
n
i
q
u
e
s,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
8
,
n
o
.
1
,
p
p
.
2
3
8
–
2
5
1
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
1
5
.
2
4
5
8
8
5
8
.
[
4
4
]
T.
Z
h
u
,
Y
.
L
i
n
,
a
n
d
Y
.
L
i
u
,
“
S
y
n
t
h
e
t
i
c
mi
n
o
r
i
t
y
o
v
e
r
sam
p
l
i
n
g
t
e
c
h
n
i
q
u
e
f
o
r
mu
l
t
i
c
l
a
s
s i
m
b
a
l
a
n
c
e
p
r
o
b
l
e
ms,”
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
7
2
,
p
p
.
3
2
7
–
3
4
0
,
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
c
o
g
.
2
0
1
7
.
0
7
.
0
2
4
.
[
4
5
]
H
.
A
.
K
h
o
r
s
h
i
d
i
a
n
d
U
.
A
i
c
k
e
l
i
n
,
“
A
s
y
n
t
h
e
t
i
c
o
v
e
r
-
sam
p
l
i
n
g
w
i
t
h
t
h
e
m
i
n
o
r
i
t
y
a
n
d
m
a
j
o
r
i
t
y
c
l
a
ss
e
s
f
o
r
i
m
b
a
l
a
n
c
e
p
r
o
b
l
e
ms,”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
s
i
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
p
p
.
1
–
1
2
,
2
0
2
0
.
[
4
6
]
Y
.
F
u
e
t
a
l
.
,
“
L
o
n
g
-
t
a
i
l
e
d
v
i
s
u
a
l
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
d
e
e
p
m
o
d
e
l
s
:
a
m
e
t
h
o
d
o
l
o
g
i
c
a
l
su
r
v
e
y
a
n
d
e
v
a
l
u
a
t
i
o
n
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
5
0
9
,
p
p
.
2
9
0
–
3
0
9
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
2
2
.
0
8
.
0
3
1
.
[
4
7
]
Y
.
S
u
n
,
M
.
S
.
K
a
m
e
l
,
A
.
K
.
C
.
W
o
n
g
,
a
n
d
Y
.
W
a
n
g
,
“
C
o
s
t
-
sen
si
t
i
v
e
b
o
o
st
i
n
g
f
o
r
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
i
m
b
a
l
a
n
c
e
d
d
a
t
a
,
”
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
4
0
,
n
o
.
1
2
,
p
p
.
3
3
5
8
–
3
3
7
8
,
2
0
0
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
c
o
g
.
2
0
0
7
.
0
4
.
0
0
9
.
[
4
8
]
T.
Y
.
L
i
n
,
P
.
G
o
y
a
l
,
R
.
G
i
r
sh
i
c
k
,
K
.
H
e
,
a
n
d
P
.
D
o
l
l
a
r
,
“
F
o
c
a
l
l
o
ss
f
o
r
d
e
n
s
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
2
,
n
o
.
2
,
p
p
.
3
1
8
–
3
2
7
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
1
8
.
2
8
5
8
8
2
6
.
[
4
9
]
Y
.
C
u
i
,
M
.
Ji
a
,
T.
Y
.
Li
n
,
Y
.
S
o
n
g
,
a
n
d
S
.
B
e
l
o
n
g
i
e
,
“
C
l
a
ss
-
b
a
l
a
n
c
e
d
l
o
ss
b
a
s
e
d
o
n
e
f
f
e
c
t
i
v
e
n
u
m
b
e
r
o
f
sam
p
l
e
s
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
ren
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
2
0
1
9
,
v
o
l
.
2
0
1
9
-
Ju
n
e
,
p
p
.
9
2
6
0
–
9
2
6
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
9
.
0
0
9
4
9
.
[
5
0
]
Z.
D
e
n
g
,
H
.
Li
u
,
Y
.
W
a
n
g
,
C
.
W
a
n
g
,
Z.
Y
u
,
a
n
d
X
.
S
u
n
,
“
P
M
L:
p
r
o
g
r
e
s
si
v
e
mar
g
i
n
l
o
ss
f
o
r
l
o
n
g
-
t
a
i
l
e
d
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
2
0
2
1
,
p
p
.
1
0
4
9
8
–
1
0
5
0
7
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
6
4
3
7
.
2
0
2
1
.
0
1
0
3
6
.
[
5
1
]
M
.
H
a
v
a
e
i
e
t
a
l
.
,
“
B
r
a
i
n
t
u
m
o
r
se
g
m
e
n
t
a
t
i
o
n
w
i
t
h
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
Me
d
i
c
a
l
i
m
a
g
e
a
n
a
l
y
s
i
s
,
p
p
.
1
8
–
3
1
,
2
0
1
7
.
[
5
2
]
O
.
S
a
g
i
a
n
d
L
.
R
o
k
a
c
h
,
“
En
s
e
m
b
l
e
l
e
a
r
n
i
n
g
:
a
s
u
r
v
e
y
,
”
Wi
l
e
y
I
n
t
e
rd
i
sc
i
p
l
i
n
a
r
y
Re
v
i
e
w
s:
D
a
t
a
Mi
n
i
n
g
a
n
d
K
n
o
w
l
e
d
g
e
D
i
s
c
o
v
e
r
y
,
v
o
l
.
8
,
n
o
.
4
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
2
/
w
i
d
m.1
2
4
9
.
[
5
3
]
N
.
V
C
h
a
w
l
a
,
A
.
L
a
z
a
r
e
v
i
c
,
L.
O
.
H
a
l
l
,
a
n
d
K
.
W
.
B
o
w
y
e
r
,
“
S
M
O
TE
B
o
o
st
:
i
m
p
r
o
v
i
n
g
p
r
e
d
i
c
t
i
o
n
o
f
t
h
e
mi
n
o
r
i
t
y
c
l
a
ss
i
n
b
o
o
s
t
i
n
g
,
”
L
e
c
t
u
r
e
N
o
t
e
s
i
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
S
u
b
ser
i
e
s
o
f
L
e
c
t
u
r
e
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
)
,
v
o
l
.
2
8
3
8
,
p
p
.
1
0
7
–
1
1
9
,
2
0
0
3
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
5
4
0
-
3
9
8
0
4
-
2
_
1
2
.
[
5
4
]
C
.
S
e
i
f
f
e
r
t
,
T.
M
.
K
h
o
s
h
g
o
f
t
a
a
r
,
J.
V
a
n
H
u
l
s
e
,
a
n
d
A
.
N
a
p
o
l
i
t
a
n
o
,
“
R
U
S
B
o
o
st
:
a
h
y
b
r
i
d
a
p
p
r
o
a
c
h
t
o
a
l
l
e
v
i
a
t
i
n
g
c
l
a
ss
i
m
b
a
l
a
n
c
e
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
S
y
st
e
m
s,
Ma
n
,
a
n
d
C
y
b
e
r
n
e
t
i
c
s
P
a
rt
A:
S
y
st
e
m
s
a
n
d
H
u
m
a
n
s
,
v
o
l
.
4
0
,
n
o
.
1
,
p
p
.
1
8
5
–
1
9
7
,
2
0
1
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TS
M
C
A
.
2
0
0
9
.
2
0
2
9
5
5
9
.
[
5
5
]
M
.
Lv
,
Y
.
R
e
n
,
a
n
d
Y
.
C
h
e
n
,
“
R
e
s
e
a
r
c
h
o
n
i
m
b
a
l
a
n
c
e
d
d
a
t
a
:
b
b
a
s
e
d
o
n
S
M
O
T
E
-
a
d
a
b
o
o
st
a
l
g
o
r
i
t
h
m
,
”
i
n
2
0
1
9
I
EE
E
3
r
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
El
e
c
t
ro
n
i
c
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
ri
n
g
,
EI
T
C
E
2
0
1
9
,
2
0
1
9
,
p
p
.
1
1
6
5
–
1
1
7
0
,
d
o
i
:
1
0
.
1
1
0
9
/
EI
TC
E4
7
2
6
3
.
2
0
1
9
.
9
0
9
4
8
5
9
.
[
5
6
]
E.
I
l
e
b
e
r
i
,
Y
.
S
u
n
,
a
n
d
Z.
W
a
n
g
,
“
P
e
r
f
o
r
ma
n
c
e
e
v
a
l
u
a
t
i
o
n
o
f
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
f
o
r
c
r
e
d
i
t
c
a
r
d
f
r
a
u
d
d
e
t
e
c
t
i
o
n
u
si
n
g
S
M
O
TE
a
n
d
A
d
a
B
o
o
s
t
,
”
I
EEE
Ac
c
e
s
s
,
v
o
l
.
9
,
p
p
.
1
6
5
2
8
6
–
1
6
5
2
9
4
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
1
3
4
3
3
0
.
[
5
7
]
Q
.
S
u
e
t
a
l
.
,
“
Ta
r
g
e
t
i
n
t
e
n
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
mo
d
e
l
b
a
s
e
d
o
n
S
M
O
TE
-
A
d
a
B
o
o
st
u
n
d
e
r
u
n
b
a
l
a
n
c
e
d
sam
p
l
e
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
2
0
2
2
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
U
n
m
a
n
n
e
d
S
y
s
t
e
m
s,
I
C
U
S
2
0
2
2
,
2
0
2
2
,
p
p
.
2
2
7
–
2
3
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
U
S
5
5
5
1
3
.
2
0
2
2
.
9
9
8
7
1
9
7
.
[
5
8
]
J.
E
d
w
a
r
d
,
M
.
M
.
R
o
s
l
i
,
a
n
d
A
.
S
e
m
a
n
,
“
A
n
e
w
mu
l
t
i
-
c
l
a
ss
r
e
b
a
l
a
n
c
i
n
g
f
r
a
mew
o
r
k
f
o
r
i
m
b
a
l
a
n
c
e
m
e
d
i
c
a
l
d
a
t
a
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
9
2
8
5
7
–
9
2
8
7
4
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
3
.
3
3
0
9
7
3
2
.
[
5
9
]
X
.
G
a
o
,
N
.
Jam
i
l
,
M
.
I
.
R
a
m
l
i
,
a
n
d
S
.
M
.
Z.
S
.
Z.
A
r
i
f
f
i
n
,
“
A
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
c
o
m
b
i
n
a
t
i
o
n
o
f
C
N
N
-
b
a
se
d
mo
d
e
l
s
w
i
t
h
e
n
s
e
mb
l
e
l
e
a
r
n
i
n
g
o
n
i
m
b
a
l
a
n
c
e
d
d
a
t
a
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
n
I
n
f
o
rm
a
t
i
c
s
Vi
su
a
l
i
z
a
t
i
o
n
,
v
o
l
.
8
,
n
o
.
1
,
p
p
.
4
5
6
–
4
6
4
,
2
0
2
4
,
d
o
i
:
1
0
.
6
2
5
2
7
/
j
o
i
v
.
8
.
1
.
2
1
9
4
.
[
6
0
]
Y
.
Tu
,
J.
F
e
n
g
,
a
n
d
Y
.
Y
a
n
g
,
“
A
A
G
:
se
l
f
-
su
p
e
r
v
i
se
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
b
y
a
u
x
i
l
i
a
r
y
a
u
g
m
e
n
t
a
t
i
o
n
w
i
t
h
G
N
T
-
X
e
n
t
l
o
s
s,”
a
rXi
v
p
re
p
r
i
n
t
a
rXi
v
:
2
0
0
9
.
0
7
9
9
4
,
2
0
2
0
.
[
6
1
]
Y
.
T
i
a
n
e
t
a
l
.
,
“
C
o
n
st
r
a
i
n
e
d
c
o
n
t
r
a
s
t
i
v
e
d
i
st
r
i
b
u
t
i
o
n
l
e
a
r
n
i
n
g
f
o
r
u
n
su
p
e
r
v
i
se
d
a
n
o
ma
l
y
d
e
t
e
c
t
i
o
n
a
n
d
l
o
c
a
l
i
s
a
t
i
o
n
i
n
m
e
d
i
c
a
l
i
ma
g
e
s,
”
i
n
Me
d
i
c
a
l
I
m
a
g
e
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
p
u
t
e
r
Ass
i
st
e
d
I
n
t
e
rve
n
t
i
o
n
–
MIC
C
AI
2
0
2
1
–
L
e
c
t
u
re
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
1
2
9
0
5
,
S
p
r
i
n
g
e
r
,
C
h
a
m,
2
0
2
1
,
p
p
.
1
2
8
–
1
4
0
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
8
7
2
4
0
-
3
_
1
3
.
[
6
2
]
Y
.
G
a
o
e
t
a
l
.
,
“
D
i
sC
o
:
r
e
me
d
y
i
n
g
s
e
l
f
-
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
o
n
l
i
g
h
t
w
e
i
g
h
t
m
o
d
e
l
s
w
i
t
h
d
i
st
i
l
l
e
d
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
,
”
L
e
c
t
u
r
e
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
su
b
ser
i
e
s
L
e
c
t
u
re
N
o
t
e
s i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
r
e
N
o
t
e
s
i
n
B
i
o
i
n
f
o
rm
a
t
i
c
s)
,
v
o
l
.
1
3
6
8
6
LN
C
S
,
p
p
.
2
3
7
–
2
5
3
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
1
9
8
0
9
-
0
_
1
4
.
[
6
3
]
H
.
R
.
Y
a
o
,
N
.
C
a
o
,
K
.
R
u
sse
l
l
,
D
.
C
.
C
h
a
n
g
,
O
.
F
r
i
e
d
e
r
,
a
n
d
J.
T
.
F
i
n
e
m
a
n
,
“
S
e
l
f
-
s
u
p
e
r
v
i
se
d
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
o
n
e
l
e
c
t
r
o
n
i
c
h
e
a
l
t
h
r
e
c
o
r
d
s
w
i
t
h
g
r
a
p
h
K
e
r
n
e
l
i
n
f
o
m
a
x
,
”
AC
M
T
r
a
n
sa
c
t
i
o
n
s
o
n
C
o
m
p
u
t
i
n
g
f
o
r
H
e
a
l
t
h
c
a
re
,
v
o
l
.
5
,
n
o
.
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
4
8
6
9
5
.
[
6
4
]
N
.
K
a
n
g
,
H
.
C
h
a
n
g
,
B
.
M
a
,
a
n
d
S
.
S
h
a
n
,
“
A
c
o
m
p
r
e
h
e
n
si
v
e
f
r
a
m
e
w
o
r
k
f
o
r
l
o
n
g
-
t
a
i
l
e
d
l
e
a
r
n
i
n
g
v
i
a
p
r
e
t
r
a
i
n
i
n
g
a
n
d
n
o
r
m
a
l
i
z
a
t
i
o
n
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
u
ra
l
N
e
t
w
o
rk
s
a
n
d
L
e
a
rn
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
5
,
n
o
.
3
,
p
p
.
3
4
3
7
–
3
4
4
9
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS.
2
0
2
2
.
3
1
9
2
4
7
5
.
[
6
5
]
J.
Tr
ä
u
b
l
e
,
L.
H
i
sc
o
x
,
C
.
Jo
h
n
so
n
,
C
.
-
B
.
S
c
h
ö
n
l
i
e
b
,
G
.
K
.
S
c
h
i
e
r
l
e
,
a
n
d
A
.
A
v
i
l
e
s
-
R
i
v
e
r
o
,
“
C
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
w
i
t
h
d
y
n
a
m
i
c
l
o
c
a
l
i
z
e
d
r
e
p
u
l
si
o
n
f
o
r
b
r
a
i
n
a
g
e
p
r
e
d
i
c
t
i
o
n
o
n
3
D
st
i
f
f
n
e
ss
ma
p
s,
”
a
rXi
v
p
r
e
p
r
i
n
t
a
rX
i
v
:
2
4
0
8
.
0
0
5
2
7
,
2
0
2
4
.
[
6
6
]
Z.
J
i
a
n
g
,
T
.
C
h
e
n
,
T.
C
h
e
n
,
a
n
d
Z.
W
a
n
g
,
“
I
mp
r
o
v
i
n
g
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
o
n
i
m
b
a
l
a
n
c
e
d
se
e
d
d
a
t
a
v
i
a
o
p
e
n
-
w
o
r
l
d
sam
p
l
i
n
g
,
”
Ad
v
a
n
c
e
s i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ssi
n
g
S
y
st
e
m
s
,
v
o
l
.
8
,
p
p
.
5
9
9
7
–
6
0
0
9
,
2
0
2
1
.
[
6
7
]
Z.
Y
a
n
g
,
J.
W
a
n
g
,
a
n
d
Y
.
Z
h
u
,
“
F
e
w
-
sh
o
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
w
i
t
h
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
,
”
i
n
E
u
r
o
p
e
a
n
c
o
n
f
e
re
n
c
e
o
n
c
o
m
p
u
t
e
r
v
i
s
i
o
n
,
2
0
2
2
,
p
p
.
2
9
3
–
3
0
9
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
2
0
0
4
4
-
1
_
1
7
.
[
6
8
]
A
.
M
a
j
e
e
,
S
.
K
o
t
h
a
w
a
d
e
,
K
.
K
i
l
l
i
a
ms
e
t
t
y
,
a
n
d
R
.
I
y
e
r
,
“
S
C
o
R
e
:
s
u
b
mo
d
u
l
a
r
c
o
mb
i
n
a
t
o
r
i
a
l
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
f
o
r
r
e
a
l
-
w
o
r
l
d
c
l
a
ss
-
i
mb
a
l
a
n
c
e
d
s
e
t
t
i
n
g
s,”
a
r
Xi
v
p
re
p
ri
n
t
a
r
Xi
v
:
2
3
1
0
.
0
0
1
6
5
.
,
2
0
2
3
.
[
6
9
]
Z.
W
a
n
g
,
C
.
P
e
n
g
,
Y
.
Z
h
a
n
g
,
N
.
W
a
n
g
,
a
n
d
L.
L
u
o
,
“
F
u
l
l
y
c
o
n
v
o
l
u
t
i
o
n
a
l
si
a
m
e
se
n
e
t
w
o
r
k
s
b
a
s
e
d
c
h
a
n
g
e
d
e
t
e
c
t
i
o
n
f
o
r
o
p
t
i
c
a
l
a
e
r
i
a
l
i
ma
g
e
s
w
i
t
h
f
o
c
a
l
c
o
n
t
r
a
s
t
i
v
e
l
o
ss,”
N
e
u
r
o
c
o
m
p
u
t
i
n
g
,
v
o
l
.
4
5
7
,
p
p
.
1
5
5
–
1
6
7
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
2
1
.
0
6
.
0
5
9
.
[
7
0
]
F
.
A
l
e
n
e
z
i
,
Ş
.
Ö
z
t
ü
r
k
,
A
.
A
r
m
g
h
a
n
,
a
n
d
K
.
P
o
l
a
t
,
“
A
n
e
f
f
e
c
t
i
v
e
h
a
s
h
i
n
g
m
e
t
h
o
d
u
s
i
n
g
W
-
S
h
a
p
e
d
c
o
n
t
r
a
s
t
i
v
e
l
o
s
s
f
o
r
i
mb
a
l
a
n
c
e
d
d
a
t
a
se
t
s,
”
E
x
p
e
rt
S
y
st
e
m
s w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
0
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
2
.
1
1
7
6
1
2
.
[
7
1
]
Y
.
Zh
a
n
g
,
B
.
H
o
o
i
,
D
.
H
u
,
J.
L
i
a
n
g
,
a
n
d
J.
F
e
n
g
,
“
U
n
l
e
a
sh
i
n
g
t
h
e
p
o
w
e
r
o
f
c
o
n
t
r
a
s
t
i
v
e
se
l
f
-
su
p
e
r
v
i
se
d
v
i
su
a
l
m
o
d
e
l
s v
i
a
c
o
n
t
r
a
s
t
-
r
e
g
u
l
a
r
i
z
e
d
f
i
n
e
-
t
u
n
i
n
g
,
”
Ad
v
a
n
c
e
s
i
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
6
,
p
p
.
2
9
8
4
8
–
2
9
8
6
0
,
2
0
2
1
.
[
7
2
]
A
.
A
u
d
i
b
e
r
t
,
A
.
G
a
u
f
f
r
e
,
a
n
d
M
.
-
R
.
A
mi
n
i
,
“
E
x
p
l
o
r
i
n
g
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
f
o
r
l
o
n
g
-
t
a
i
l
e
d
mu
l
t
i
-
l
a
b
e
l
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
J
o
i
n
t
Eu
r
o
p
e
a
n
C
o
n
f
e
re
n
c
e
o
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
a
n
d
K
n
o
w
l
e
d
g
e
D
i
sc
o
v
e
ry
i
n
D
a
t
a
b
a
ses
,
p
p
.
2
4
5
–
2
6
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
7
0
3
6
8
-
3
_
1
5
.
[
7
3
]
M
.
R
.
H
.
Ta
h
e
r
,
F
.
H
a
g
h
i
g
h
i
,
M
.
B
.
G
o
t
w
a
y
,
a
n
d
J
.
Li
a
n
g
,
“
C
A
i
D
:
c
o
n
t
e
x
t
-
a
w
a
r
e
i
n
s
t
a
n
c
e
d
i
s
c
r
i
m
i
n
a
t
i
o
n
f
o
r
se
l
f
-
su
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
i
n
me
d
i
c
a
l
i
m
a
g
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
R
e
se
a
rc
h
,
2
0
2
2
,
v
o
l
.
1
7
2
,
p
p
.
5
3
5
–
5
5
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.