I
nte
rna
t
io
na
l J
o
urna
l o
f
Rec
o
nfig
ura
ble a
nd
E
m
bedd
e
d
Sy
s
t
em
s
(
I
J
RE
S)
Vo
l.
10
,
No
.
1
,
Ma
r
ch
2
0
2
1
,
p
p
.
1
~
10
I
SS
N
: 2
0
8
9
-
4
8
6
4
,
DOI
: 1
0
.
1
1
5
9
1
/ijre
s
.
v
1
0
.
i1
.
p
p
1
-
10
1
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
r
es.ia
esco
r
e.
co
m
A
d
esig
n
m
e
thodo
lo
g
y
f
o
r
a
ppro
x
i
ma
te
m
ulti
pliers i
n
co
nv
o
lutiona
l
n
eura
l
n
etworks
:
A
c
a
se o
f
M
NIST
K
ent
a
Sh
ira
n
e
,
T
a
k
a
hiro
Y
a
m
a
mo
t
o
,
H
iro
y
u
k
i To
m
iy
a
ma
G
ra
d
u
a
te S
c
h
o
o
l
o
f
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
,
Rit
s
u
m
e
ik
a
n
Un
i
v
e
r
sity
,
Ja
p
a
n
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
10
,
2
0
20
R
ev
is
ed
J
an
1
0
,
2
0
21
Acc
ep
ted
J
an
20
,
2
0
21
In
th
is
p
a
p
e
r,
we
p
re
se
n
t
a
c
a
se
stu
d
y
o
n
a
p
p
r
o
x
ima
te
m
u
lt
ip
li
e
rs
f
o
r
M
NIST
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
(
CNN
)
.
We
a
p
p
l
y
a
p
p
ro
x
ima
te
m
u
lt
ip
li
e
rs
with
d
if
fe
re
n
t
b
i
t
-
wid
t
h
t
o
th
e
c
o
n
v
o
l
u
ti
o
n
lay
e
r
in
M
NIST
C
NN
,
e
v
a
lu
a
te
th
e
a
c
c
u
ra
c
y
o
f
M
NIS
T
c
las
sifica
ti
o
n
,
a
n
d
a
n
a
l
y
z
e
th
e
trad
e
-
o
f
f
b
e
twe
e
n
a
p
p
ro
x
ima
te
m
u
lt
ip
l
ier’s
a
re
a
,
c
ri
ti
c
a
l
p
a
th
d
e
lay
a
n
d
t
h
e
a
c
c
u
ra
c
y
.
Ba
se
d
o
n
th
e
re
su
lt
s
o
f
t
h
e
e
v
a
l
u
a
ti
o
n
a
n
d
a
n
a
ly
sis,
we
p
ro
p
o
se
a
d
e
sig
n
m
e
th
o
d
o
l
o
g
y
fo
r
a
p
p
r
o
x
ima
te
m
u
lt
i
p
li
e
rs.
T
h
e
a
p
p
ro
x
ima
te
m
u
lt
ip
li
e
r
s
c
o
n
si
st
o
f
so
m
e
p
a
rti
a
l
p
ro
d
u
c
ts,
wh
ich
a
re
c
a
re
fu
ll
y
se
lec
ted
a
c
c
o
rd
i
n
g
t
o
th
e
C
NN
in
p
u
t.
Wi
th
th
is
m
e
th
o
d
o
lo
g
y
,
we
fu
rth
e
r
re
d
u
c
e
th
e
a
re
a
a
n
d
th
e
d
e
lay
o
f
th
e
m
u
lt
ip
li
e
rs wit
h
k
e
e
p
in
g
h
ig
h
a
c
c
u
ra
c
y
o
f
th
e
M
NI
S
T
c
las
sifica
ti
o
n
.
K
ey
w
o
r
d
s
:
Ap
p
r
o
x
im
ate
c
o
m
p
u
tin
g
Ap
p
r
o
x
im
ate
m
u
ltip
lier
C
NN
MN
I
ST
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
:
Ken
ta
Sh
ir
an
e
Gr
ad
u
ate
Sch
o
o
l
o
f
Scien
ce
a
n
d
E
n
g
in
ee
r
in
g
R
its
u
m
eik
an
Un
iv
er
s
ity
1
-
1
-
1
No
ji
-
Hig
ash
i,
Ku
s
atsu
,
Sh
ig
a
5
2
5
-
8
5
7
7
,
J
ap
an
E
m
ail:
k
en
ta.
s
h
ir
an
e@
to
m
iy
a
m
a
-
lab
.
o
r
g
1.
I
NT
RO
D
UCT
I
O
N
C
o
n
v
o
lu
tio
n
al
Neu
r
al
Netwo
r
k
s
(
C
NNs)
ar
e
well
-
k
n
o
wn
as
o
n
e
o
f
t
h
e
s
tate
-
of
-
th
e
-
ar
t
ap
p
r
o
ac
h
es
in
th
e
f
ield
s
o
f
m
ed
ia
p
r
o
ce
s
s
in
g
s
u
ch
as
co
m
p
u
ter
v
is
io
n
,
s
p
e
ec
h
r
ec
o
g
n
itio
n
,
a
n
d
s
o
o
n
.
R
ec
en
tly
,
d
esig
n
s
o
f
co
m
p
u
tin
g
s
y
s
tem
s
with
C
NNs
h
av
e
b
ee
n
ex
ten
s
iv
ely
s
tu
d
ied
b
y
r
esear
ch
er
s
.
I
n
g
en
er
al,
C
NNs
ar
e
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
d
u
e
to
r
e
p
etitio
n
o
f
th
e
m
atr
ix
m
u
ltip
licatio
n
th
at
is
th
e
m
o
s
t
c
r
itical
b
o
ttlen
ec
k
i
n
C
NNs
[
1
]
.
I
n
o
r
d
e
r
to
r
ed
u
ce
t
h
e
co
m
p
u
tatio
n
al
co
s
t
o
f
m
atr
i
x
m
u
ltip
licatio
n
,
ap
p
r
o
x
im
ate
co
m
p
u
tin
g
attr
ac
ts
in
cr
ea
s
in
g
atten
tio
n
.
Ap
p
r
o
x
i
m
ate
co
m
p
u
tin
g
is
co
n
s
id
er
e
d
as
a
p
r
o
m
is
in
g
a
p
p
r
o
ac
h
t
o
d
esig
n
o
f
ar
ea
-
,
p
er
f
o
r
m
an
ce
-
o
r
p
o
wer
-
ef
f
ic
ien
t
cir
cu
its
f
o
r
ap
p
licatio
n
s
wh
ich
ar
e
r
esis
tan
t
to
a
ce
r
tain
am
o
u
n
t
o
f
co
m
p
u
tatio
n
al
in
ac
cu
r
ac
y
.
Ap
p
r
o
x
im
ate
ca
lcu
latio
n
cir
cu
its
h
av
e
b
ee
n
ac
tiv
el
y
s
u
r
v
ey
ed
[
2
-
5
]
.
Acc
o
r
d
in
g
to
th
e
wo
r
k
in
[
6
]
an
d
[
7
]
,
wh
i
ch
in
v
esti
g
ated
th
e
r
ec
en
t
s
tu
d
ies
in
th
e
ar
ea
o
f
ap
p
r
o
x
im
ate
co
m
p
u
tin
g
,
th
e
d
esig
n
o
f
ap
p
r
o
x
im
ate
m
u
lti
p
lier
s
h
as
ex
ten
s
iv
ely
b
ee
n
s
tu
d
ied
.
Ma
n
y
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
p
r
o
p
o
s
ed
ca
lcu
late
th
e
u
p
p
er
b
its
ex
ac
tly
an
d
lo
wer
b
its
ap
p
r
o
x
im
ately
[
8
-
1
0
]
.
Yam
am
o
to
et
al.
p
r
o
p
o
s
ed
a
m
eth
o
d
o
l
o
g
y
to
s
y
s
tem
atica
ll
y
d
esig
n
8
-
b
it
a
p
p
r
o
x
im
ate
a
r
r
ay
m
u
ltip
lier
s
b
y
g
r
a
d
u
ally
r
e
m
o
v
in
g
ad
d
er
s
,
a
n
d
an
aly
ze
d
th
e
tr
ad
e
-
o
f
f
s
am
o
n
g
cir
cu
it
ar
ea
,
d
elay
,
p
o
wer
a
n
d
ac
cu
r
ac
y
o
f
th
e
a
p
p
r
o
x
im
a
te
m
u
ltip
lier
s
[
1
1
]
.
B
o
r
o
u
m
an
d
et
al.
ex
p
l
o
r
ed
t
h
e
d
esig
n
s
p
ac
e
o
f
ap
p
r
o
x
i
m
ate
m
u
ltip
lier
s
u
s
in
g
a
f
am
ily
o
f
ap
p
r
o
x
im
ate
co
m
p
r
ess
o
r
s
as
b
u
ild
in
g
b
lo
c
k
s
f
o
r
th
e
p
a
r
tial
p
r
o
d
u
ct
r
ed
u
ctio
n
tr
ee
[
1
2
]
.
T
h
e
y
p
r
ese
n
te
d
a
to
o
l
th
at
allo
ws
th
e
u
s
er
to
s
p
ec
if
y
an
allo
wa
b
le
lev
el
o
f
er
r
o
r
to
ler
a
n
ce
,
an
d
r
etu
r
n
s
th
e
m
in
im
u
m
a
r
e
a,
d
elay
,
o
r
p
o
wer
ap
p
r
o
x
im
ate
m
u
ltip
lier
th
at
p
r
o
v
id
es
th
e
lev
el
o
f
ac
c
u
r
ac
y
.
L
iu
et
al.
p
r
o
p
o
s
ed
a
n
ew
a
p
p
r
o
x
im
ate
m
u
ltip
lier
with
h
ig
h
ly
ef
f
ici
en
t
p
ar
tial
p
r
o
d
u
ct
ac
c
u
m
u
latio
n
a
n
d
er
r
o
r
r
ed
u
ctio
n
[
1
3
]
.
T
h
e
n
ew
ap
p
r
o
x
im
ate
m
u
ltip
lier
h
as
h
ig
h
ac
cu
r
ac
y
b
u
t
is
m
o
r
e
ef
f
icien
t
in
p
o
wer
an
d
p
er
f
o
r
m
an
ce
.
I
n
t
h
ese
wo
r
k
s
,
ap
p
r
o
x
im
ate
ca
lcu
latio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4
8
6
4
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
1
–
10
2
cir
cu
its
ar
e
d
esig
n
ed
with
s
m
all
ca
lcu
latio
n
er
r
o
r
.
I
n
ad
d
it
io
n
,
th
er
e
ar
e
m
an
y
r
esear
ch
es
o
n
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
th
at
r
ed
u
ce
t
h
e
er
r
o
r
in
th
e
f
in
al
o
u
tp
u
t o
f
C
NN.
As
m
en
tio
n
ed
ab
o
v
e,
ap
p
r
o
x
i
m
ate
co
m
p
u
tin
g
is
s
tu
d
ied
as
an
ef
f
ec
tiv
e
m
eth
o
d
s
p
ec
ializ
ed
to
C
N
N
[
1
4
-
1
6
]
.
An
e
r
r
o
r
r
esil
ien
ce
a
n
aly
s
is
was
p
er
f
o
r
m
e
d
i
n
o
r
d
er
to
d
eter
m
in
e
k
ey
co
n
s
tr
ain
ts
f
o
r
th
e
d
esig
n
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
th
at
ar
e
em
p
l
o
y
ed
in
th
e
r
esu
ltin
g
s
tr
u
ctu
r
e
o
f
C
NN
[
1
7
]
.
M
o
r
eo
v
e
r
,
th
e
p
a
p
er
s
h
o
wed
th
e
ca
p
ab
ilit
y
o
f
th
e
b
ac
k
-
p
r
o
p
ag
atio
n
lear
n
i
n
g
a
lg
o
r
ith
m
f
o
r
C
NNs
co
n
tain
in
g
th
e
ap
p
r
o
x
im
at
e
m
u
ltip
lier
s
in
MN
I
ST
an
d
Str
ee
t
-
View
Ho
u
s
e
Nu
m
b
er
d
ata
s
et.
Ham
m
ad
et
al.
s
im
u
lated
th
e
VGGN
et
[
1
8
]
wh
ich
em
p
lo
y
s
an
ap
p
r
o
x
i
m
ate
m
u
ltip
lier
in
th
e
lay
e
r
s
[
1
9
]
.
T
h
e
s
im
u
latio
n
r
esu
lts
s
h
o
w
th
at
th
e
ap
p
r
o
x
im
ate
m
u
ltip
lier
k
ee
p
s
t
h
e
h
ig
h
ac
cu
r
ac
y
o
f
VGGN
et.
Sim
et
al.
p
r
o
p
o
s
ed
a
n
ew
Sto
ch
asti
c
C
o
m
p
u
tin
g
(
SC
)
m
u
ltip
licatio
n
alg
o
r
ith
m
an
d
its
v
ec
to
r
ex
ten
s
io
n
wh
ich
is
ca
lled
S
C
-
MV
M
(
Ma
tr
ix
-
Vec
to
r
Mu
ltip
lier
)
[
2
0
]
.
T
h
e
ex
p
er
im
e
n
tal
r
esu
lts
s
h
o
w
th
at
n
ew
SC
-
b
ased
C
NN
is
m
o
r
e
ac
c
u
r
ate
an
d
4
0
×
to
4
9
0
×
m
o
r
e
en
e
r
g
y
-
e
f
f
icien
t
in
co
m
p
u
tatio
n
th
an
th
e
c
o
n
v
e
n
tio
n
al
SC
-
b
ased
o
n
es.
SC
is
o
n
e
o
f
th
e
m
eth
o
d
s
o
f
ap
p
r
o
x
i
m
ate
co
m
p
u
tin
g
an
d
h
as
b
ee
n
ac
tiv
ely
s
tu
d
ied
[
2
1
,
2
2
]
.
C
o
u
r
b
ar
i
au
x
et
al.
p
r
o
p
o
s
ed
a
m
eth
o
d
wh
ich
tr
ain
s
n
eu
r
al
n
etwo
r
k
p
ar
am
eter
s
as
b
in
ar
y
weig
h
t
an
d
ac
tiv
ati
o
n
[
2
3
]
.
T
h
is
n
etwo
r
k
,
wh
ich
is
ca
lled
B
in
ar
y
Net,
ca
n
m
ak
e
a
r
ed
u
ctio
n
o
f
m
em
o
r
y
u
s
ag
e
an
d
u
s
e
XNOR
cir
cu
it
s
as
a
m
u
ltip
lier
,
s
o
m
an
y
ap
p
r
o
ac
h
es
o
f
ac
ce
ler
atio
n
o
r
/an
d
im
p
lem
en
tatio
n
h
av
e
b
ee
n
p
r
o
p
o
s
ed
[
2
4
-
2
7
]
.
L
i
et
al.
an
d
B
alaji
et
al.
d
esig
n
e
d
8
-
b
it
f
i
x
ed
-
p
o
in
t
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
L
eNe
t
an
d
im
p
lem
en
ted
to
FP
GA
[
2
8
,
2
9
]
.
T
h
ese
ar
e
m
o
r
e
ef
f
icien
t
in
d
elay
an
d
r
eso
u
r
ce
th
a
n
ex
ac
t
c
o
m
p
u
tatio
n
with
h
i
g
h
ac
c
u
r
ac
y
o
f
MN
I
ST
class
if
icatio
n
.
Fo
r
th
ese
wo
r
k
s
,
ap
p
r
o
x
im
ate
co
m
p
u
tin
g
ca
n
r
e
d
u
ce
c
o
m
p
u
t
atio
n
al
r
eso
u
r
ce
s
an
d
C
NN’
s
tr
ain
ed
p
a
r
am
eter
s
.
I
n
th
is
p
ap
er
,
we
p
r
esen
t
a
d
e
s
i
g
n
m
eth
o
d
o
l
o
g
y
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
MN
I
ST
C
NN.
As
a
p
r
elim
in
ar
y
p
r
ep
ar
atio
n
o
f
d
ei
g
n
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
,
we
d
esig
n
6
4
m
u
ltip
lier
s
with
s
im
p
le
r
ed
u
ctio
n
o
f
th
e
b
it
-
wid
th
o
f
an
8
-
b
it
m
u
ltip
lier
,
an
d
ev
alu
ate
tr
a
d
e
-
o
f
f
b
etwe
en
th
e
ac
cu
r
ac
y
o
f
M
NI
ST
class
if
icat
io
n
,
th
e
ar
ea
an
d
th
e
d
ela
y
o
f
t
h
e
m
u
ltip
lier
s
.
B
ased
o
n
th
e
an
al
y
s
is
,
we
d
esig
n
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
b
ased
o
n
two
m
eth
o
d
s
.
On
e
o
f
th
e
m
eth
o
d
s
is
b
ased
o
n
co
m
p
r
eh
en
s
iv
e
an
aly
s
is
o
f
u
p
p
er
2
-
b
it
p
ar
ti
al
p
r
o
d
u
cts
o
f
8
-
b
it
m
u
ltip
lier
,
an
d
th
e
o
th
e
r
is
s
elec
tio
n
o
f
th
e
p
ar
tial
p
r
o
d
u
c
ts
ac
co
r
d
in
g
to
r
esu
lt
o
f
f
ir
s
t
m
eth
o
d
an
d
8
-
b
it
m
u
ltip
lier
’
s
an
aly
s
is
.
W
ith
th
ese
im
p
r
o
v
e
m
en
ts
,
we
f
u
r
th
e
r
r
ed
u
ce
th
e
ar
ea
an
d
th
e
d
ela
y
o
f
th
e
m
u
ltip
lier
s
with
k
ee
p
in
g
h
ig
h
ac
cu
r
ac
y
in
MN
I
ST
clas
s
if
icatio
n
.
T
h
e
r
est o
f
th
is
p
ap
er
is
o
r
g
an
ized
as f
o
llo
ws.
Sectio
n
2
.
s
h
o
ws
o
u
tlin
e
o
f
g
e
n
er
al
C
NN
an
d
MN
I
ST
C
NN
in
th
is
p
ap
er
.
Sectio
n
3
d
escr
ib
es
s
im
p
le
ap
p
r
o
ac
h
to
b
it
-
wid
th
r
ed
u
ctio
n
an
d
t
h
e
ex
p
e
r
im
en
ts
.
Sectio
n
4
.
d
escr
ib
es
a
d
esig
n
m
eth
o
d
o
lo
g
y
o
f
ap
p
r
o
x
im
ate
m
u
lt
ip
lie
r
with
two
-
s
tep
ap
p
r
o
ac
h
,
an
d
S
ec
tio
n
5
.
c
o
n
clu
d
es th
is
p
a
p
er
with
a
s
u
m
m
ar
y
.
2.
CNN
F
O
R
M
NI
ST
C
L
A
SS
I
F
I
CAT
I
O
N
I
n
th
is
s
ec
tio
n
,
we
ex
p
lain
a
g
en
er
al
C
o
n
v
o
lu
tio
n
al
Ne
u
r
al
Netwo
r
k
(
C
NN)
,
its
co
n
v
o
lu
tio
n
lay
er
an
d
MN
I
ST
C
NN
in
th
is
wo
r
k
.
C
NN
is
a
n
eu
r
al
n
e
two
r
k
wh
ich
h
as
d
ee
p
lay
e
r
in
clu
d
i
n
g
co
n
v
o
lu
tio
n
la
y
er
.
C
NN
is
k
n
o
wn
as
a
h
ig
h
-
ac
cu
r
ac
y
im
ag
e
class
if
icatio
n
m
o
d
el.
On
e
o
f
th
e
m
o
s
t
p
o
p
u
lar
i
m
ag
e
class
if
icatio
n
m
o
d
els
is
Alex
n
et
[
3
0
]
.
I
n
2
0
1
2
,
Alex
n
et
wo
n
I
L
SVR
C
b
y
a
b
ig
m
a
r
g
in
o
v
er
o
th
e
r
m
o
d
els.
T
h
is
v
ic
to
r
y
tr
ig
g
er
ed
p
r
ev
alen
ce
o
f
C
NN
in
th
e
wo
r
ld
a
n
d
m
a
n
y
o
f
n
e
u
r
al
n
etwo
r
k
s
in
im
a
g
e
class
if
icatio
n
ar
e
b
ased
o
n
C
NN
to
d
ay
.
2
.
1
.
Co
nv
o
lutio
n
l
a
y
er
I
n
th
is
s
u
b
s
ec
tio
n
,
we
ex
p
lain
ab
o
u
t
a
g
e
n
er
al
co
n
v
o
lu
tio
n
lay
er
.
A
C
o
n
v
o
l
u
tio
n
lay
er
p
er
f
o
r
m
s
co
n
v
o
l
u
tio
n
o
p
er
atio
n
o
n
an
i
n
p
u
t
im
ag
e
with
a
f
ilter
(
weig
h
t)
,
its
o
u
tp
u
ts
s
u
m
u
p
b
ias,
an
d
th
e
o
u
tp
u
ts
ar
e
p
r
o
p
a
g
ated
t
o
th
e
n
e
x
t
lay
er
o
r
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
e
f
il
ter
an
d
b
ias
a
r
e
tr
ain
e
d
a
n
d
s
av
ed
in
co
n
v
o
l
u
tio
n
lay
er
.
I
n
co
n
v
o
l
u
tio
n
o
p
e
r
at
io
n
,
th
e
p
r
o
d
u
cts
o
f
in
p
u
t
an
d
f
ilter
ar
e
s
u
m
m
ed
u
p
an
d
s
to
r
ed
to
th
e
co
r
r
esp
o
n
d
in
g
o
u
tp
u
t.
C
o
n
v
o
lu
tio
n
o
p
e
r
atio
n
p
er
f
o
r
m
s
at
an
in
ter
v
al
w
h
ich
is
d
ef
in
e
d
“strid
e”
.
Fig
u
r
e
1
r
ep
r
esen
ts
an
ex
am
p
le
o
f
co
n
v
o
lu
tio
n
o
p
er
atio
n
.
W
h
en
a
5
×
5
in
p
u
t,
a
3
×3
f
iler
an
d
s
tr
id
e
v
a
lu
e
“1
”
ar
e
g
iv
e
n
,
th
e
s
ize
o
f
th
e
o
u
tp
u
t
is
o
r
g
an
ized
as
3
×3
.
I
n
th
is
ca
s
e,
th
e
co
n
v
o
l
u
tio
n
lay
er
co
n
tain
s
8
1
m
u
ltip
licatio
n
s
p
er
f
ilter
.
I
n
g
e
n
er
al,
C
NN
co
n
tai
n
s
m
an
y
f
ilter
s
.
I
n
t
h
is
wo
r
k
,
C
NN
co
n
tain
s
3
0
f
ilter
s
an
d
t
h
e
co
n
v
o
lu
tio
n
la
y
er
p
er
f
o
r
m
s
m
u
lti
p
licatio
n
o
p
er
at
io
n
s
4
3
2
,
0
0
0
tim
es p
er
im
ag
e.
Fig
u
r
e
1
.
C
o
n
v
o
lu
tio
n
lay
er
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4
8
6
4
A
d
esig
n
meth
o
d
o
lo
g
y
fo
r
a
p
p
r
o
xima
te
mu
ltip
lier
s
in
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
ks:
…
(
K
en
ta
S
h
ir
a
n
e
)
3
2
.
2
.
An
e
x
a
m
ple o
f
CNN
f
o
r
M
NIS
T
I
n
th
is
s
u
b
s
ec
tio
n
,
we
ex
p
lai
n
ab
o
u
t
MN
I
ST
C
NN
in
th
i
s
p
ap
er
.
Fig
u
r
e
2
s
h
o
ws
th
e
m
o
d
el
o
f
MN
I
ST
C
NN
an
d
MN
I
ST
d
at
aset.
MN
I
ST
C
N
N
co
n
s
is
t
s
o
f
C
o
n
v
o
lu
tio
n
(
co
n
v
o
lu
tio
n
la
y
er
)
,
Po
o
lin
g
(
m
ax
p
o
o
lin
g
lay
er
)
,
Af
f
in
e
(
af
f
in
e
lay
er
)
,
R
eL
U
an
d
So
f
t
m
ax
.
Nex
t
p
ar
t
d
e
s
cr
ib
es
ea
ch
o
f
th
e
lay
er
s
o
r
t
h
e
ac
tiv
atio
n
f
u
n
cti
o
n
s
,
ex
ce
p
t
f
r
o
m
th
e
co
n
v
o
l
u
tio
n
lay
er
.
Po
o
lin
g
s
elec
ts
th
e
m
a
x
im
u
m
n
u
m
b
er
o
f
in
p
u
t
elem
en
ts
in
f
ilter
r
an
g
e
an
d
s
to
r
es
th
e
c
o
r
r
esp
o
n
d
in
g
o
u
tp
u
t.
T
h
is
lay
er
ca
n
r
ed
u
ce
p
r
o
ce
s
s
in
g
weig
h
t w
ith
co
n
s
tan
t a
p
p
r
o
x
im
atio
n
.
T
h
is
o
p
er
ati
o
n
d
o
es n
o
t c
o
n
tain
p
ar
am
eter
s
s
u
ch
as we
ig
h
t o
r
b
ias,
an
d
k
ee
p
s
th
e
s
p
atial
in
f
o
r
m
atio
n
o
f
in
p
u
t im
ag
e.
Af
f
in
e
m
u
ltip
lies
ea
ch
in
p
u
t
elem
en
t
an
d
weig
h
t
v
alu
e,
s
u
m
s
u
p
th
ese
p
r
o
d
u
cts
an
d
s
to
r
es
th
e
co
r
r
esp
o
n
d
in
g
o
u
tp
u
t.
T
h
is
lay
er
co
n
n
ec
ts
all
s
p
atial
in
f
o
r
m
atio
n
to
ea
ch
class
.
T
h
is
o
p
er
atio
n
co
n
tain
s
p
ar
am
eter
s
wh
ich
ar
e
weig
h
t a
n
d
b
ias,
an
d
d
o
es n
o
t k
ee
p
th
e
s
p
atial
in
f
o
r
m
atio
n
o
f
in
p
u
t im
ag
e.
R
eL
U
an
d
So
f
tm
ax
ar
e
ac
tiv
a
tio
n
f
u
n
ctio
n
s
wh
ich
p
r
e
p
ar
e
i
n
p
u
t
d
ata.
R
eL
U
f
ix
es
“0
”
o
r
less
v
alu
e
to
“0
”
an
d
s
to
r
es
th
e
c
o
r
r
es
p
o
n
d
in
g
o
u
tp
u
t.
I
t
is
k
n
o
w
n
as
r
am
p
f
u
n
ctio
n
.
So
f
tm
a
x
f
ix
es
in
p
u
t
v
alu
e
to
p
r
o
b
a
b
ilit
y
v
alu
e
.
T
h
is
f
u
n
cti
o
n
is
u
s
ed
b
e
f
o
r
e
f
in
ally
o
u
t
p
u
t
an
d
h
i
g
h
est
p
r
o
b
ab
ilit
y
v
al
u
e’
s
class
is
C
NN
o
u
tp
u
t.
T
h
e
C
NN
tr
ain
ed
MN
I
ST
d
a
ta
s
et.
MN
I
ST
d
ata
s
et
co
n
ta
in
s
h
an
d
wr
itten
d
ig
its
f
r
o
m
“
0
”
to
“9
”,
wh
ich
is
a
co
llectio
n
o
f
2
8
×2
8
q
u
asi
b
in
ar
y
im
a
g
es,
an
d
th
e
p
ix
el
v
alu
e
is
v
ar
ied
f
r
o
m
“
0
”
to
“1
”.
MN
I
ST
is
o
n
e
o
f
th
e
m
o
s
t
elem
en
tar
y
d
a
tasets
in
im
ag
e
clas
s
if
icatio
n
o
f
C
NN,
s
o
we
f
ir
s
tly
u
s
e
t
h
is
d
ataset
to
ex
am
in
e
th
e
im
p
ac
t o
f
a
p
p
r
o
x
im
ate
m
u
ltip
lier
to
im
ag
e
class
if
icatio
n
.
Fig
u
r
e
2
.
An
e
x
am
p
le
o
f
C
NN
f
o
r
MN
I
ST
3.
P
RE
L
I
M
I
NAR
Y
ANA
L
YSI
S O
N
B
I
T
-
WI
D
T
H
T
h
is
s
ec
tio
n
d
escr
ib
es
h
o
w
to
r
ed
u
ce
b
it
-
wid
th
o
f
m
u
ltip
lic
atio
n
an
d
th
e
ex
p
e
r
im
en
ts
to
F
ig
u
r
e
o
u
t
th
e
s
ca
le
o
f
m
u
ltip
lier
wh
ich
C
NN
n
ee
d
s
.
T
h
e
e
x
p
er
im
e
n
t
i
s
th
e
p
r
e
p
ar
atio
n
o
f
S
ec
tio
n
4
.
an
d
th
e
p
u
r
p
o
s
e
is
an
ex
p
lo
r
atio
n
o
f
th
e
m
u
ltip
lier
wh
ich
k
ee
p
s
h
ig
h
ac
c
u
r
ac
y
o
f
im
ag
e
class
if
icatio
n
.
3
.
1
.
B
it
-
Width
r
educt
io
n
a
pp
ro
a
ch
I
n
th
is
wo
r
k
,
we
d
esig
n
d
i
f
f
er
en
t
b
it
-
wid
th
m
u
ltip
lier
s
b
y
r
ed
u
cin
g
p
ar
tial
p
r
o
d
u
c
ts
o
f
8
-
b
it
m
u
ltip
lier
.
T
h
e
8
-
b
it
m
u
ltip
lie
r
is
g
en
er
al
s
m
all
s
ca
le,
b
u
t
M
NI
ST
is
k
n
o
wn
as
d
ataset
wh
i
ch
ca
n
b
e
class
if
ied
b
y
u
s
in
g
s
m
all
s
ca
le
cir
cu
it.
T
h
e
s
tr
u
ctu
r
e
o
f
B
in
ar
y
Net
is
d
if
f
er
e
n
t
f
r
o
m
C
NN
in
th
is
wo
r
k
,
b
u
t
B
in
ar
y
Net
h
as
a
h
ig
h
ac
cu
r
a
cy
o
f
im
ag
e
class
if
icatio
n
an
d
MN
I
ST
class
if
icatio
n
d
o
es
n
o
t
s
ee
m
to
n
ee
d
h
ig
h
p
r
ec
is
io
n
m
u
ltip
licatio
n
.
T
h
e
8
-
b
it
m
u
ltip
lier
is
b
ased
o
n
an
ex
ac
t
8
-
b
it
ar
r
a
y
m
u
ltip
lier
s
in
ce
th
e
8
-
b
it
ar
r
ay
m
u
ltip
lie
r
is
s
o
s
m
all,
an
d
th
e
8
-
b
it
m
u
ltip
lier
d
o
es
n
o
t
h
av
e
t
o
ap
p
ly
W
allac
e
tr
ee
an
d
/o
r
b
o
o
th
’
s
m
u
ltip
licatio
n
b
ec
au
s
e
o
f
b
ad
tr
ad
eo
f
f
b
etwe
en
t
h
e
o
v
e
r
h
ea
d
an
d
t
h
e
p
er
f
o
r
m
a
n
ce
im
p
r
o
v
em
e
n
t
o
f
th
e
m
u
ltip
lier
.
Fig
u
r
e
3
is
an
ex
ac
t
8
-
b
it
ar
r
ay
m
u
ltip
lier
.
T
h
e
m
u
lti
p
licatio
n
u
s
es
m
u
ltip
lican
d
an
d
m
u
ltip
lier
to
o
b
tain
p
ar
tia
l
p
r
o
d
u
cts.
Su
m
o
f
th
e
p
ar
tial
p
r
o
d
u
cts
is
a
p
r
o
d
u
ct.
Su
m
o
p
e
r
atio
n
s
tar
ts
f
r
o
m
u
p
p
er
-
r
i
g
h
t
t
o
lo
wer
-
lef
t.
T
h
e
p
r
o
d
u
ct
o
f
8
-
b
it
m
u
ltip
lier
b
ec
o
m
es 1
6
-
b
it
p
r
o
d
u
ct.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4
8
6
4
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
1
–
10
4
I
n
th
is
r
ed
u
ctio
n
,
we
r
e
p
lace
th
e
m
u
ltip
lican
d
o
r
th
e
m
u
ltip
l
ier
with
“0
”
in
o
r
d
er
f
r
o
m
l
o
west
-
o
r
d
er
b
it.
B
y
th
is
o
p
er
atio
n
,
th
e
co
r
r
esp
o
n
d
in
g
m
u
ltip
lier
o
r
m
u
ltip
lican
d
is
u
s
eless
,
an
d
th
is
p
ar
tial
p
r
o
d
u
ct
’
s
v
alu
e
b
ec
o
m
es
‘
0
’
.
I
n
ad
d
itio
n
,
th
e
m
u
ltip
lican
d
an
d
m
u
ltip
lier
h
a
v
e
8
-
b
it
in
f
o
r
m
atio
n
,
an
d
we
c
an
r
ed
u
ce
d
if
f
er
en
t
b
it
-
wid
th
b
etwe
en
th
e
m
u
ltip
lican
d
an
d
th
e
m
u
ltip
lier
.
Fi
g
u
r
e
4
is
an
ex
am
p
le
o
f
7
×6
-
b
it
m
u
ltip
lier
.
W
e
r
ep
lace
least
b
it
o
f
m
u
ltip
lic
an
d
a
n
d
l
o
wer
2
-
b
it
o
f
th
e
m
u
ltip
lier
with
“
0
”,
s
o
we
c
an
d
elete
2
2
p
ar
tial
p
r
o
d
u
cts
an
d
lo
wer
3
-
b
it
o
f
p
r
o
d
u
ct.
W
e
ca
n
r
e
d
u
ce
d
if
f
er
en
t
b
it
-
wid
th
b
etwe
en
th
e
8
m
u
ltip
lican
d
s
an
d
8
m
u
ltip
lier
s
,
s
o
we
ca
n
cr
ea
te
6
4
m
u
ltip
lier
s
in
d
if
f
er
en
t b
it
-
w
id
th
.
Fig
u
r
e
3
.
An
ex
ac
t 8
-
b
it m
u
lti
p
lier
Fig
u
r
e
4
.
An
ex
am
p
le
o
f
7
×
6
-
b
it
m
u
ltip
lier
3
.
2
.
P
re
li
m
ina
ry
r
esu
lt
s
W
e
d
esig
n
ed
MN
I
ST
C
NN
in
o
r
d
er
to
a
p
p
ly
s
elf
-
m
a
d
e
m
u
ltip
lier
s
.
W
e
ap
p
lied
d
if
f
e
r
en
t
b
it
-
wid
th
m
u
ltip
lier
s
to
MN
I
ST
C
NN
a
n
d
ev
alu
ated
th
e
tr
ad
e
-
o
f
f
s
o
f
th
e
ac
cu
r
ac
y
o
f
MN
I
ST
class
if
icatio
n
,
th
e
ar
ea
an
d
th
e
d
ela
y
o
f
t
h
e
m
u
ltip
lier
.
T
h
e
ex
p
er
im
en
ts
ca
n
F
ig
u
r
e
o
u
t
th
e
s
ca
le
o
f
m
u
ltip
lier
wh
i
ch
C
NN
n
ee
d
s
,
an
d
we
im
p
r
o
v
e
th
e
m
u
ltip
lier
,
wh
ich
k
ee
p
s
h
ig
h
ac
cu
r
ac
y
o
f
i
m
ag
e
class
if
icatio
n
,
b
y
r
e
f
er
e
n
ce
to
e
x
p
er
im
e
n
tal
r
esu
lts
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
W
e
ca
n
r
ed
u
ce
d
if
f
er
e
n
t
b
i
t
-
wid
th
b
etwe
en
th
e
m
u
ltip
lican
d
an
d
m
u
ltip
lier
,
s
o
we
ca
n
cr
ea
te
6
4
m
u
ltip
lier
s
in
d
if
f
er
en
t
b
it
-
wid
th
.
W
e
u
s
e
C
NN
p
ar
am
eter
s
lear
n
ed
2
0
0
0
tr
ain
in
g
im
ag
es.
W
e
m
ea
s
u
r
e
th
e
im
ag
e
class
i
f
icatio
n
ac
cu
r
ac
y
w
h
en
C
NN
r
ec
o
g
n
ize
1
0
0
0
test
im
ag
es.
T
h
e
ac
cu
r
ac
y
wi
th
ex
ac
t
co
m
p
u
tatio
n
,
wh
ich
is
3
2
-
b
it
f
lo
atin
g
p
o
in
t
m
u
ltip
licatio
n
,
is
9
8
.
4
%.
T
h
e
d
escr
ip
ti
o
n
o
f
m
u
ltip
lier
s
is
wr
itten
in
Ver
ilo
g
.
W
e
s
y
n
th
esis
th
e
6
4
m
u
ltip
lier
s
b
y
L
e
o
n
ar
d
o
Sp
ec
tr
u
m
a
n
d
m
ea
s
u
r
e
th
e
n
u
m
b
er
o
f
g
ate
an
d
cr
itical
p
ath
d
elay
.
W
e
u
s
e
Nan
g
ate4
5
wh
ich
is
4
5
n
m
p
r
o
ce
s
s
lib
r
a
r
y
.
W
h
en
we
s
y
n
th
esis
th
e
cir
c
u
it
o
p
tim
ized
m
in
im
izin
g
ar
ea
.
Fig
u
r
e
5
(
a)
-
(
c)
ar
e
th
e
e
x
p
er
i
m
en
tal
r
esu
lts
o
f
b
it
-
wid
th
r
e
d
u
ctio
n
.
T
h
e
f
ilter
d
ata
ar
e
m
u
ltip
lican
d
an
d
th
e
im
a
g
e
d
ata
ar
e
m
u
ltip
lier
.
Fig
u
r
e
5
(
a)
is
th
e
r
esu
lt
o
f
ac
c
u
r
ac
y
,
Fig
u
r
e
5
(
b
)
is
th
e
r
esu
lt
o
f
t
h
e
ar
ea
,
an
d
Fig
u
r
e
5
(
c
)
is
th
e
r
esu
lt o
f
th
e
d
elay
.
I
n
Fig
u
r
e
5
(
a)
,
ac
cu
r
ac
y
in
8
×
8
-
b
it is
9
8
.
4
% a
n
d
ac
cu
r
ac
ies in
3
×2
-
b
it
an
d
4
×1
-
b
it
ar
e
al
m
o
s
t
th
e
s
am
e
as
th
e
ex
ac
t
o
n
e
.
T
h
e
a
cc
u
r
ac
y
i
n
2
×
1
-
b
it
is
9
7
.
1
%
a
n
d
th
e
ac
cu
r
ac
y
in
less
th
an
2
×1
-
b
it
d
ec
r
ea
s
e
s
ig
n
if
ican
tly
.
W
h
en
ev
e
r
th
e
f
ilter
b
it
-
wid
th
is
1
,
th
e
ac
c
u
r
ac
y
is
ex
tr
em
ely
lo
w
.
T
h
er
ef
o
r
e,
th
e
MN
I
ST
C
NN
n
ee
d
s
ab
o
u
t
2
×
2
-
o
r
m
o
r
e
b
it
m
u
ltip
lier
to
k
ee
p
h
ig
h
ac
cu
r
ac
y
o
f
im
ag
e
class
if
icatio
n
.
I
n
Fig
u
r
e
5
(
b
)
s
h
o
ws
th
at
th
e
ar
ea
o
f
m
u
ltip
lier
d
ec
r
ea
s
es
in
p
r
o
p
o
r
ti
o
n
to
th
e
s
ca
le
o
f
m
u
ltip
lier
.
T
h
e
a
r
ea
in
3
×2
-
b
it
is
1
5
g
ates
th
at
co
n
tain
s
HA
(
h
alf
ad
d
er
)
a
n
d
m
a
n
y
AN
D
g
ates,
th
e
ar
ea
in
4
×1
-
b
it
is
4
g
ates
wh
ich
ar
e
4
AND
g
ates
an
d
th
e
o
n
e
in
2
×1
-
b
it
is
2
g
ates
wh
ic
h
ar
e
two
AND
g
ates.
I
n
Fig
u
r
e
5
(
c
)
s
h
o
ws
th
at
th
e
d
e
lay
o
f
m
u
ltip
lier
d
ec
r
ea
s
es
in
p
r
o
p
o
r
tio
n
t
o
th
e
s
ca
le
o
f
m
u
lt
ip
lier
.
T
h
e
d
ela
y
in
3
×2
-
b
it is
0
.
1
0
n
s
.
T
h
e
d
elay
in
4
×1
-
b
it is
0
.
0
2
n
s
as we
ll a
s
th
e
2
×1
-
b
it.
Partial Pro
du
ct
+
Pr
od
uct
+
Partial Pro
du
ct
Pr
od
uct
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4
8
6
4
A
d
esig
n
meth
o
d
o
lo
g
y
fo
r
a
p
p
r
o
xima
te
mu
ltip
lier
s
in
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
ks:
…
(
K
en
ta
S
h
ir
a
n
e
)
5
T
ab
le
1
is
th
e
d
etail
o
f
u
p
p
e
r
o
r
d
er
3
-
b
it
im
ag
e
b
it
-
wid
th
.
C
ases
o
f
3
×2
-
b
it
an
d
4
×1
-
b
it
ar
e
s
am
e
ac
cu
r
ac
y
an
d
co
n
tain
f
iv
e
p
a
r
tial
p
r
o
d
u
cts.
Ho
wev
er
,
th
e
ar
ea
o
f
th
e
3
×2
-
b
it
m
u
ltip
lier
is
lar
g
er
th
an
th
at
o
f
4
×1
-
b
it,
an
d
th
e
d
ela
y
o
f
th
e
3
×2
-
b
it
m
u
ltip
lier
is
lo
n
g
er
d
elay
th
an
a
ca
s
e
o
f
4
×1
-
b
it.
T
h
er
ef
o
r
e
,
a
ca
s
e
o
f
3
×2
-
b
it
is
b
etter
p
o
wer
ef
f
icien
cy
th
an
a
ca
s
e
o
f
4
×1
-
b
i
t.
B
esid
es,
f
ilter
d
ata
(
m
u
ltip
lican
d
)
a
r
e
m
o
r
e
im
p
o
r
tan
t
to
MN
I
ST
C
NN
th
an
im
ag
e
d
ata
(
m
u
ltip
lier
)
.
I
n
th
e
ex
p
er
im
en
tal
r
esu
lts
,
M
NI
ST
C
NN
n
ee
d
s
4
g
ates
to
ac
h
iev
e
alm
o
s
t
th
e
s
am
e
ac
cu
r
ac
y
as
ac
cu
r
ac
y
o
f
cl
ass
if
icatio
n
with
ex
ac
t
c
o
m
p
u
tatio
n
.
I
n
a
d
d
itio
n
,
wh
en
th
e
n
u
m
b
e
r
o
f
g
ates
is
3
in
3
×1
-
b
it,
ac
cu
r
ac
y
is
9
8
.
0
%
an
d
th
e
MN
I
ST
C
NN
n
ee
d
s
ab
o
u
t
3
×2
,
4
×1
o
r
more
-
b
it
m
u
ltip
lier
to
k
ee
p
h
i
g
h
ac
c
u
r
ac
y
o
f
im
ag
e
class
if
ic
atio
n
.
T
h
e
r
ef
o
r
e,
in
th
e
n
ex
t
s
ec
tio
n
,
th
e
p
r
o
p
o
s
e
d
d
esig
n
m
eth
o
d
o
lo
g
y
is
u
s
ed
to
ac
h
iev
e
h
ig
h
er
ac
cu
r
ac
y
with
3
o
r
f
ewe
r
g
ates.
(
a)
(
b
)
(
c)
Fig
u
r
e
5
.
E
x
p
er
im
e
n
tal
r
esu
lts
o
f
b
it
-
w
id
th
r
ed
u
ctio
n
(
a)
Acc
u
r
ac
y
,
(
b
)
Ar
ea
,
(
c)
Dela
y
T
ab
le
1
.
Deta
il o
f
u
p
p
e
r
o
r
d
er
3
-
b
it im
ag
e
b
it
-
wid
th
F
i
l
t
e
r
b
i
t
-
w
i
d
t
h
I
mag
e
b
i
t
-
w
i
d
t
h
A
c
c
u
r
a
c
y
(
%)
A
r
e
a
(
g
a
t
e
s
)
D
e
l
a
y
(
n
s)
1
1
2
0
.
4
1
0
.
0
2
2
1
9
7
.
1
2
0
.
0
2
3
1
9
8
.
0
3
0
.
0
2
4
1
9
8
.
5
4
0
.
0
2
5
1
9
8
.
5
5
0
.
0
2
6
1
9
8
.
4
6
0
.
0
2
7
1
9
8
.
4
7
0
.
0
2
8
1
9
8
.
4
9
0
.
0
2
1
2
2
2
.
1
2
0
.
0
2
2
2
9
7
.
6
7
0
.
0
9
3
2
9
8
.
5
15
0
.
1
0
4
2
9
8
.
7
22
0
.
2
0
5
2
9
8
.
6
28
0
.
2
6
6
2
9
8
.
6
35
0
.
3
3
7
2
9
8
.
6
42
0
.
3
9
8
2
9
8
.
6
48
0
.
4
6
1
3
2
3
.
2
3
0
.
0
2
2
3
9
7
.
9
17
0
.
1
5
3
3
9
8
.
5
30
0
.
2
8
4
3
9
8
.
6
42
0
.
3
5
5
3
9
8
.
6
55
0
.
4
2
6
3
9
8
.
3
67
0
.
4
8
7
3
9
8
.
3
80
0
.
5
5
8
3
9
8
.
3
92
0
.
6
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4
8
6
4
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
1
–
10
6
4.
A
DE
S
I
G
N
M
E
T
H
O
DO
L
O
G
Y
F
O
R
AP
P
RO
XIM
A
T
E
M
UL
T
I
P
L
I
E
R
S
T
h
is
s
ec
tio
n
d
escr
ib
es
a
d
esig
n
m
eth
o
d
o
lo
g
y
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
MN
I
ST
C
NN.
I
n
th
e
f
ir
s
t
s
u
b
s
ec
tio
n
,
we
ex
p
lain
an
o
v
er
v
iew
o
f
a
d
esig
n
m
et
h
o
d
o
lo
g
y
o
f
a
p
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
C
NN.
I
n
th
e
n
ex
t
two
s
u
b
s
ec
tio
n
s
,
we
a
p
p
l
ied
th
e
d
esig
n
m
et
h
o
d
o
lo
g
y
t
o
MN
I
ST
C
NN.
At
ev
er
y
s
te
p
,
we
e
v
alu
ated
th
e
tr
ad
e
-
o
f
f
s
o
f
th
e
ac
c
u
r
ac
y
o
f
MN
I
ST
class
if
icatio
n
,
th
e
ar
ea
an
d
th
e
d
elay
o
f
th
e
m
u
ltip
lier
.
4
.
1
.
A
d
esig
n
m
et
ho
do
lo
g
y
T
h
is
s
u
b
s
ec
tio
n
d
escr
ib
es
o
v
e
r
v
iew
o
f
t
h
e
p
r
o
p
o
s
ed
d
esig
n
m
eth
o
d
o
l
o
g
y
o
f
a
p
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
C
NN.
Fig
u
r
e
6
is
f
lo
wch
ar
t
o
f
th
e
p
r
o
p
o
s
ed
d
esig
n
m
eth
o
d
o
lo
g
y
.
I
n
th
e
f
ir
s
t
h
ar
f
o
f
th
e
m
eth
o
d
o
lo
g
y
,
we
an
aly
ze
p
ar
tial
p
r
o
d
u
cts
o
f
ex
ac
t
m
u
ltip
lier
an
d
s
elec
t
s
ig
n
if
ican
t
p
ar
tial
p
r
o
d
u
cts
f
o
r
C
NN
as
an
ap
p
r
o
x
im
ate
m
u
ltip
lier
.
I
f
th
e
ac
c
u
r
ac
y
o
f
C
NN
is
n
o
t
s
ig
n
if
ican
tly
h
ig
h
er
o
r
th
e
ar
ea
o
r
d
ela
y
d
o
n
’
t
s
atis
f
y
u
s
er
r
eq
u
ir
em
e
n
ts
,
we
an
aly
ze
a
b
r
o
ad
er
s
co
p
e
ag
ain
.
I
n
th
e
s
ec
o
n
d
h
ar
f
o
f
t
h
e
m
eth
o
d
o
lo
g
y
,
we
a
n
aly
ze
i
n
m
o
r
e
d
etail
p
ar
tial
p
r
o
d
u
cts
o
f
ex
ac
t
m
u
ltip
lier
f
o
r
c
o
m
b
in
i
n
g
to
th
e
ap
p
r
o
x
im
ate
m
u
ltip
lier
d
esig
n
ed
in
th
e
f
ir
s
t
h
alf
o
f
th
e
m
eth
o
d
o
lo
g
y
.
I
f
th
e
ac
cu
r
ac
y
o
f
C
NN
is
lo
wer
th
an
ex
ac
t
o
n
e,
we
an
aly
ze
a
b
r
o
ad
er
s
co
p
e
ag
ain
.
I
n
t
h
e
d
esig
n
m
et
h
o
d
o
lo
g
y
,
we
ca
n
d
esig
n
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
th
at
m
ain
tain
a
h
ig
h
r
ec
o
g
n
itio
n
r
ate
o
f
C
NN
with
o
u
t
co
m
p
r
o
m
is
in
g
th
e
u
s
er
'
s
r
eq
u
ir
em
e
n
ts
.
I
n
th
e
n
ex
t two
s
u
b
s
ec
tio
n
s
,
we
ap
p
lied
th
e
d
e
s
ig
n
m
eth
o
d
o
lo
g
y
to
MN
I
ST
C
NN.
Fig
u
r
e
6
.
Flo
wch
a
r
t
o
f
th
e
p
r
o
p
o
s
ed
d
esig
n
m
et
h
o
d
o
lo
g
y
4
.
2
.
Co
m
prehens
iv
e
a
na
ly
s
is
o
f
2
-
bit
p
a
rt
ia
l
p
ro
du
ct
s
T
h
is
s
u
b
s
ec
tio
n
d
escr
ib
es
th
e
ex
am
p
le
o
f
th
e
co
m
p
r
e
h
en
s
iv
e
an
aly
s
is
o
f
2
×2
-
b
it
m
u
ltip
lier
.
Acc
o
r
d
in
g
to
th
e
ex
p
e
r
im
en
tal
r
esu
lts
o
f
S
ec
tio
n
3
,
we
an
aly
ze
d
o
n
ly
h
ig
h
e
r
2
-
b
it p
ar
tial p
r
o
d
u
cts.
Fig
u
r
e
7
is
an
ex
ac
t
2
×2
-
b
it
m
u
ltip
lier
.
‘
a
’
is
f
ilter
b
it,
‘
b
’
is
im
ag
e
b
it,
‘
p
’
is
p
r
o
d
u
ct,
an
d
HA
is
h
al
f
-
ad
d
er
cir
cu
it.
W
e
r
ed
u
ce
d
p
ar
tial
p
r
o
d
u
cts
f
r
o
m
lo
wer
b
it
in
S
ec
tio
n
3
,
b
u
t
we
s
elec
ted
o
n
e
to
f
o
u
r
p
a
r
tial
p
r
o
d
u
cts
in
o
u
r
Sta
rt
C
o
m
pr
e
he
nsi
ve
ana
ly
si
s
o
f
N
×
N
-
bit
m
ulti
plie
r
T
h
e
ar
e
a
an
d
de
la
y
sat
isf
y
u
ser
re
qu
ir
e
m
e
n
t
s
Se
le
c
t
(N+
1
)
-
th
up
p
e
rmo
st
p
art
ia
l
p
ro
d
uc
ts
o
f
the
app
ro
xim
at
e
m
ulti
p
lie
r
with co
m
p
re
he
nsi
ve
ana
ly
si
s
o
f
M
-
th
up
p
e
rmo
st
p
art
ia
l
p
ro
d
uc
t
Se
le
c
t
si
gnific
ant
partia
l
pr
o
duct
s
as
an appr
o
xim
at
e
m
ulti
plie
r
F
inis
h
T
h
e
acc
u
ra
c
y
of
C
N
N
is
sig
n
if
ican
t
ly
h
ig
h
e
r
T
h
e
C
NN
acc
u
ra
c
y
is
low
e
r
t
h
an
e
x
act
on
e
No
Y
e
s
M
=
M
+1
N
=
N+
1
N
=
1
M
=
1
Y
e
s
No
Y
e
s
No
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4
8
6
4
A
d
esig
n
meth
o
d
o
lo
g
y
fo
r
a
p
p
r
o
xima
te
mu
ltip
lier
s
in
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
ks:
…
(
K
en
ta
S
h
ir
a
n
e
)
7
d
esig
n
m
eth
o
d
o
lo
g
y
.
W
e
an
aly
ze
d
th
e
ac
cu
r
ac
y
o
f
MN
I
ST
class
if
icat
io
n
,
th
e
ar
ea
a
n
d
th
e
d
elay
o
f
th
e
ap
p
r
o
x
im
ate
m
u
ltip
lier
wh
ich
h
as
o
n
e
to
f
o
u
r
p
ar
tial
p
r
o
d
u
cts
o
f
2
×2
-
b
it
m
u
ltip
lier
co
m
p
r
eh
en
s
iv
ely
.
I
n
th
is
ex
p
er
im
en
t,
we
ca
n
ex
p
l
o
r
e
th
e
co
m
b
in
ati
o
n
o
f
p
ar
tial
p
r
o
d
u
cts
wh
ich
k
ee
p
s
h
ig
h
ac
cu
r
ac
y
o
f
im
ag
e
class
if
icatio
n
.
T
h
e
ex
p
er
im
e
n
tal
co
n
d
itio
n
s
ar
e
b
ased
o
n
S
ec
tio
n
3
.
T
ab
le
2
is
th
e
r
esu
lt
o
f
co
m
p
r
eh
en
s
iv
e
a
n
aly
s
is
o
f
t
h
e
2
-
b
i
t
p
ar
tial
p
r
o
d
u
cts.
W
h
en
th
e
n
u
m
b
er
o
f
p
ar
tial
p
r
o
d
u
cts
is
1
,
th
e
ca
s
e
o
f
a
7
b
6
is
th
e
h
ig
h
est
ac
cu
r
ac
y
,
b
u
t
it
is
n
o
t
en
o
u
g
h
h
ig
h
a
cc
u
r
ac
y
.
W
h
en
th
e
n
u
m
b
er
o
f
p
a
r
tial
p
r
o
d
u
cts
is
2
,
th
e
ca
s
e
o
f
a
6
b
7
an
d
a
7
b
7
is
h
ig
h
est
ac
cu
r
ac
y
,
9
7
.
1
%.
I
n
th
is
ca
s
e,
th
e
n
u
m
b
er
o
f
g
ates
is
2
an
d
d
elay
is
0
.
0
2
n
s
.
W
h
en
th
e
n
u
m
b
e
r
o
f
p
a
r
tial
p
r
o
d
u
cts
is
3
o
r
4
,
n
o
o
t
h
er
ca
s
e
o
f
p
ar
tial
p
r
o
d
u
cts
is
m
o
r
e
ef
f
icien
cy
th
an
th
e
ca
s
e
o
f
a
6
b
7
an
d
a
7
b
7
.
F
ig
u
r
e
8
is
th
e
l
o
g
ic
lev
el
cir
cu
it
o
f
a
7
b
7
an
d
a
6
b
7
.
HAs
ar
e
r
em
o
v
ed
.
I
n
th
e
s
ec
o
n
d
h
ar
f
o
f
th
e
m
et
h
o
d
o
lo
g
y
,
we
tr
y
to
ac
h
iev
e
h
i
g
h
er
ac
c
u
r
a
cy
b
y
a
6
b
7
,
a
7
b
7
an
d
an
o
th
er
p
ar
tial p
r
o
d
u
ct.
Fig
u
r
e
7
.
2
-
b
it e
x
ac
t
m
u
ltip
lier
T
ab
le
2
.
C
o
m
p
r
eh
en
s
iv
e
a
n
aly
s
is
in
u
p
p
er
o
r
d
er
2
-
b
it
p
a
r
tial
p
r
o
d
u
cts
N
u
mb
e
r
o
f
P
a
r
t
i
a
l
P
r
o
d
u
c
t
P
a
r
t
i
a
l
P
r
o
d
u
c
t
A
c
c
u
r
a
c
y
(
%)
A
r
e
a
(
g
a
t
e
s
)
D
e
l
a
y
(
n
s)
1
a
6
b
6
4
3
.
7
1
0
.
0
2
a
6
b
7
1
7
.
8
1
0
.
0
2
a
7
b
6
6
3
.
6
1
0
.
0
2
a
7
b
7
2
0
.
4
1
0
.
0
2
2
a
6
b
6
, a
6
b
7
6
9
.
5
2
0
.
0
2
a
6
b
6
, a
7
b
6
8
6
.
6
2
0
.
0
2
a
6
b
6
, a
7
b
7
9
4
.
9
2
0
.
0
2
a
6
b
7
, a
7
b
6
9
1
.
0
4
0
.
0
6
a
6
b
7
, a
7
b
7
9
7
.
1
2
0
.
0
2
a
7
b
6
, a
7
b
7
2
2
.
1
2
0
.
0
2
3
a
6
b
6
, a
6
b
7
,
a
7
b
6
9
0
.
5
5
0
.
0
8
a
6
b
6
, a
6
b
7
,
a
7
b
7
9
0
.
0
3
0
.
0
2
a
6
b
6
, a
7
b
6
,
a
7
b
7
9
6
.
9
3
0
.
0
2
a
6
b
7
, a
7
b
6
,
a
7
b
7
9
7
.
5
6
0
.
0
8
4
a
6
b
6
, a
6
b
7
,
a
7
b
6
,
a
7
b
7
9
7
.
6
7
0
.
0
9
Fig
u
r
e
8
.
L
o
g
ic
-
lev
el
cir
c
u
it o
f
a
7
b
7
an
d
a
6
b
7
4
.
3
.
Select
io
n o
f
t
he
t
hird
u
pp
er
m
o
s
t
p
ro
du
ct
T
h
is
s
u
b
s
ec
tio
n
d
escr
ib
es
th
e
ex
am
p
le
o
f
s
elec
tio
n
o
f
th
e
th
ir
d
u
p
p
er
m
o
s
t
p
r
o
d
u
cts.
Fig
u
r
e
9
is
a
lo
g
ic
-
lev
el
cir
c
u
it
o
f
a
6
b
7
,
a
7
b
7
an
d
a
i
b
j
.
a
i
b
j
is
o
n
e
o
f
th
e
te
n
p
ar
tial
p
r
o
d
u
cts
wh
ic
h
is
p
o
s
s
ib
le
to
o
u
tp
u
t
u
p
p
er
th
an
p
11
wh
ic
h
is
th
e
f
if
t
h
u
p
p
er
m
o
s
t
p
r
o
d
u
ct.
I
n
th
e
t
en
ca
s
es,
we
an
aly
ze
th
e
ac
cu
r
ac
y
o
f
MN
I
ST
class
if
icatio
n
,
th
e
ar
ea
an
d
th
e
d
elay
o
f
th
e
m
u
ltip
lier
.
T
h
e
e
x
p
er
im
en
tal
co
n
d
itio
n
s
ar
e
b
ased
o
n
Sectio
n
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4
8
6
4
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
1
–
10
8
T
ab
le
3
is
th
e
r
esu
lt
o
f
s
elec
tio
n
o
f
t
h
e
th
i
r
d
s
ig
n
if
ican
t
p
r
o
d
u
ct.
T
h
e
n
u
m
b
er
o
f
g
ates
an
d
th
e
d
ela
y
alwa
y
s
k
ee
p
3
g
ates
an
d
0
.
0
2
n
s
,
r
esp
ec
tiv
ely
.
T
h
e
h
ig
h
est
ac
cu
r
ac
y
ac
h
iev
es
9
8
.
4
%
in
a
5
b
6
,
an
d
th
is
is
th
e
s
am
e
r
esu
lt
as
ex
ac
t
co
m
p
u
tatio
n
.
T
h
er
ef
o
r
e,
we
ac
h
iev
e
t
h
e
s
am
e
ac
cu
r
ac
y
as
MN
I
ST
class
if
icat
io
n
with
ex
ac
t c
o
m
p
u
tatio
n
b
y
u
s
in
g
a
p
p
r
o
x
im
ate
m
u
ltip
lier
co
m
p
o
s
e
d
o
f
a
7
b
7
, a
6
b
7
an
d
a
5
b
6
.
Fig
u
r
e
1
0
is
th
e
lo
g
ic
-
lev
el
ci
r
cu
it
o
f
a
6
b
7
,
a
7
b
7
a
n
d
a
5
b
6
.
I
n
th
e
ca
s
e
o
f
MN
I
ST
C
NN,
th
e
p
r
o
p
o
s
ed
d
esig
n
m
eth
o
d
o
r
o
g
y
is
ef
f
icie
n
t
f
o
r
d
esig
n
i
n
g
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
with
s
m
all
s
ca
le,
l
o
w
d
elay
an
d
h
ig
h
im
ag
e
r
ec
o
g
n
itio
n
o
f
C
NN.
B
ec
au
s
e
a
6
b
7
an
d
a
7
b
7
c
o
n
tain
n
eith
er
a
5
n
o
r
b
6
,
we
c
o
n
s
id
er
t
h
at
m
an
y
v
ar
i
eties
o
f
m
u
ltip
lican
d
an
d
m
u
ltip
lier
im
p
r
o
v
e
C
NN
ac
cu
r
ac
y
o
f
cl
ass
if
icatio
n
.
I
n
ad
d
itio
n
,
a
5
b
6
i
s
u
s
u
ally
o
u
tp
u
t
t
o
p
11
,
b
u
t
it
is
o
u
tp
u
t
to
p
12
in
th
is
ca
s
e.
T
h
er
f
o
r
e,
we
co
n
s
id
er
ed
th
at
h
ig
h
e
r
b
it
in
f
o
r
m
atio
n
is
n
o
t
alwa
y
s
m
o
r
e
im
p
o
r
tan
t to
im
a
g
e
class
if
icatio
n
with
C
NN
th
an
lo
wer
b
it in
f
o
r
m
atio
n
.
Fig
u
r
e
9
.
L
o
g
ic
-
lev
el
cir
c
u
it o
f
a
7
b
7
,
a
6
b
7
an
d
a
i
b
j
T
ab
le
3
.
Selectio
n
o
f
th
e
th
i
r
d
s
ig
n
if
ican
t p
r
o
d
u
ct
p
12
A
c
c
u
r
a
c
y
(
%)
A
r
e
a
(
g
a
t
e
s
)
D
e
l
a
y
(
n
s)
a
4
b
7
9
8
.
2
3
0
.
0
2
a
5
b
6
9
8
.
4
3
0
.
0
2
a
5
b
7
98
.0
3
0
.
0
2
a
6
b
5
9
6
.
6
3
0
.
0
2
a
6
b
6
9
6
.
9
3
0
.
0
2
a
6
b
7
9
6
.
2
3
0
.
0
2
a
7
b
4
9
7
.
3
3
0
.
0
2
a
7
b
5
9
7
.
4
3
0
.
0
2
a
7
b
6
9
7
.
3
3
0
.
0
2
a
7
b
7
9
7
.
3
3
0
.
0
2
Fig
u
r
e
10
.
L
o
g
ic
-
le
v
el
cir
cu
it
o
f
a
7
b
7
,
a
6
b
7
an
d
a
5
b
6
4
.
4
.
Co
m
pa
riso
n
T
ab
le
4
is
a
c
o
m
p
ar
is
o
n
tab
le
o
f
m
u
ltip
lier
s
with
s
im
p
le
b
it
-
wid
th
r
e
d
u
ctio
n
an
d
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
b
ased
o
n
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
.
T
h
e
la
b
els
8
×8
,
3
×
2
,
2
×
2
an
d
2
×1
a
r
e
b
ased
o
n
an
aly
s
is
o
f
Sectio
n
3
th
e
lab
el
AP
(
a
p
p
r
o
x
im
ate)
2
×
1
is
b
ased
o
n
s
u
b
s
e
ctio
n
4
.
2
,
an
d
t
h
e
lab
el
A
P
3
×2
is
b
ased
o
n
s
u
b
s
ec
tio
n
4
.
3
.
AP 3
×2
-
b
it m
u
ltip
lier
h
as th
e
s
am
e
ac
cu
r
ac
y
o
f
MN
I
ST
r
ec
o
g
n
itio
n
as th
e
e
x
ac
t 8
×8
an
d
3
×
2
-
b
it
m
u
ltip
lier
s
.
I
n
ad
d
itio
n
,
th
e
ar
ea
o
f
AP
3
×2
-
b
it
m
u
ltip
lier
im
p
r
o
v
e
d
b
y
8
0
%
o
v
er
th
e
a
r
ea
o
f
ex
ac
t
3
×2
-
b
it
m
u
ltip
lier
,
an
d
th
e
d
elay
o
f
A
P
3
×2
-
b
it
m
u
ltip
lier
im
p
r
o
v
ed
b
y
8
0
%
o
v
er
th
e
ar
ea
o
f
ex
ac
t
3
×2
-
b
it
m
u
ltip
lier
.
T
h
er
ef
o
r
e,
th
e
d
esig
n
m
eth
o
d
o
lo
g
y
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
is
ef
f
icien
t f
o
r
MN
I
ST
C
NN.
T
ab
le
4
.
C
o
m
p
a
r
is
o
n
tab
le
o
f
m
u
ltip
lier
s
with
s
im
p
le
b
it
-
wid
th
r
ed
u
ctio
n
an
d
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
M
u
l
t
i
p
l
i
e
r
A
c
c
u
r
a
c
y
(
%)
A
r
e
a
(
g
a
t
e
s
)
D
e
l
a
y
(
n
s)
8
×
8
9
8
.
4
3
1
2
1
.
2
6
3
×
2
9
8
.
5
15
0
.
1
0
2
×
2
97
.
6
7
0
.
0
9
2
×
1
9
7
.
1
2
0
.
0
2
A
P
2
×
1
(
a
7
b
7
a
n
d
a
6
b
7
)
9
7
.
1
2
0
.
0
2
A
P
3
×
2
(a
6
b
7
, a
7
b
7
a
n
d
a
5
b
6
.)
9
8
.
4
3
0
.
0
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4
8
6
4
A
d
esig
n
meth
o
d
o
lo
g
y
fo
r
a
p
p
r
o
xima
te
mu
ltip
lier
s
in
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
ks:
…
(
K
en
ta
S
h
ir
a
n
e
)
9
5.
CO
NCLU
SI
O
NS
I
n
th
is
p
a
p
er
,
we
p
r
o
p
o
s
ed
a
d
esig
n
m
eth
o
d
o
lo
g
y
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
f
o
r
C
NN.
W
e
ac
h
iev
e
d
h
ig
h
ac
cu
r
ac
y
o
f
MN
I
ST
class
if
icatio
n
with
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
b
ased
o
n
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
.
W
ith
th
e
d
esig
n
m
eth
o
d
o
lo
g
y
,
b
y
u
s
in
g
t
h
e
ap
p
r
o
x
im
ate
m
u
ltip
lier
,
we
ac
h
ie
v
ed
th
e
s
am
e
ac
cu
r
ac
y
as
MN
I
ST
class
if
icatio
n
with
ex
ac
t
co
m
p
u
tatio
n
a
n
d
im
p
r
o
v
e
d
th
e
ar
ea
a
n
d
d
elay
p
er
f
o
r
m
an
ce
.
I
n
f
u
tu
r
e,
we
p
lan
to
ap
p
ly
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
o
f
ap
p
r
o
x
im
ate
m
u
ltip
lier
s
to
m
o
r
e
co
m
p
lex
an
d
lar
g
er
s
ca
le
C
NN.
B
esid
es,
we
p
lan
to
im
p
lem
en
t a
p
p
r
o
x
im
ated
C
NN
o
n
FP
GA
an
d
p
lan
t
o
an
aly
ze
th
e
im
p
ac
t o
n
FP
GA.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
wo
r
k
is
in
p
ar
t su
p
p
o
r
ted
b
y
KAKE
NHI
1
9
H0
4
0
8
1
an
d
2
0
H0
0
5
9
0
.
RE
F
E
R
E
NC
E
S
[1
]
K.
G
u
o
,
S
.
Zen
g
,
J.
Yu
,
Y.
Wan
g
a
n
d
H.
Ya
n
g
,
“
A
su
rv
e
y
o
f
F
P
G
A
b
a
se
d
n
e
u
ra
l
n
e
two
rk
a
c
c
e
lera
to
r
,
”
ACM
T
ra
n
sa
c
ti
o
n
s
o
n
Rec
o
n
f
ig
u
ra
b
le
T
e
c
h
n
o
l
o
g
y
a
n
d
S
y
ste
ms
,
v
o
l.
1
,
n
o
.
1
,
2
0
1
8
.
[2
]
F
.
M
.
d
e
l
Ca
m
p
o
,
A.
M
o
ra
les
-
Re
y
e
s,
R.
P
e
re
z
-
An
d
ra
d
e
,
R.
Cu
m
p
li
d
o
,
A.
G
.
Oro
z
c
o
-
Lu
g
o
a
n
d
C
.
F
e
re
g
rin
o
,
“
A
m
u
lt
i
-
c
y
c
le
fi
x
e
d
p
o
in
t
sq
u
a
re
ro
o
t
m
o
d
u
le
fo
r
F
P
G
As
,
”
IEI
CE
E
lec
tro
n
ics
Exp
re
ss
,
v
o
l
.
9
,
n
o.
1
1
,
p
p
.
9
7
1
-
9
7
7
,
2
0
1
2
.
[3
]
D.
S
h
i
n
a
n
d
S
.
K.
G
u
p
ta,
“
Ap
p
ro
x
ima
te
lo
g
ic
sy
n
t
h
e
sis
fo
r
e
rro
r
to
l
e
ra
n
t
a
p
p
li
c
a
ti
o
n
s
,
”
2
0
1
0
De
sig
n
,
Au
t
o
ma
ti
o
n
&
T
e
st i
n
E
u
ro
p
e
Co
n
fer
e
n
c
e
&
Exh
ib
it
i
o
n
(DA
T
E
2
0
1
0
)
,
2
0
1
0
.
[4
]
N.
Zh
u
,
W.
L
.
G
o
h
a
n
d
K
.
S
.
Y
e
o
,
“
An
e
n
h
a
n
c
e
d
lo
w
-
p
o
we
r
h
ig
h
-
sp
e
e
d
a
d
d
e
r
f
o
r
e
rro
r
-
t
o
lera
n
t
a
p
p
li
c
a
ti
o
n
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
1
2
t
h
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
siu
m
o
n
In
te
g
ra
ted
Ci
rc
u
it
s
(
IS
IC'0
9
)
,
2
0
0
9
,
p
p
.
6
9
-
72
.
[5
]
V.
G
u
p
ta,
D.
M
o
h
a
p
a
tra,
A.
Ra
g
h
u
n
a
th
a
n
a
n
d
K.
R
o
y
,
“
Lo
w
-
p
o
we
r
d
ig
it
a
l
sig
n
a
l
p
r
o
c
e
ss
in
g
u
si
n
g
a
p
p
r
o
x
ima
te
a
d
d
e
rs,”
IEE
E
T
r
a
n
s
a
c
ti
o
n
s
o
n
C
o
mp
u
ter
-
Ai
d
e
d
De
sig
n
o
f
I
n
teg
r
a
t
e
d
Circ
u
it
s
a
n
d
S
y
ste
ms
,
v
o
l.
32
,
n
o
.
1
,
p
p
.
1
2
4
-
1
3
7
,
2
0
1
3
.
[6
]
J.
Ha
n
a
n
d
M
.
Ors
h
a
n
s
k
y
,
“
Ap
p
r
o
x
ima
te
c
o
m
p
u
ti
n
g
:
An
e
m
e
rg
in
g
p
a
ra
d
ig
m
fo
r
e
n
e
rg
y
-
e
fficie
n
t
d
e
sig
n
,
”
i
n
2
0
1
3
1
8
t
h
IEE
E
Eu
r
o
p
e
a
n
T
e
st
S
y
mp
o
s
iu
m
(ET
S
)
,
2
0
1
3
,
p
p
.
1
-
6
.
[7
]
Q.
Xu
,
N.
Kim
a
n
d
T.
M
y
t
k
o
wic
z
,
“
Ap
p
r
o
x
ima
te
c
o
m
p
u
ti
n
g
:
A
su
r
v
e
y
,
”
IEE
E
De
sig
n
&
T
e
st
,
v
o
l
.
3
3
,
n
o
.
1
,
p
p
.
8
-
2
2
,
2
0
1
5
.
[8
]
B.
S
h
a
o
,
a
n
d
P
.
Li
,
“
Arra
y
-
b
a
se
d
a
p
p
ro
x
ima
te arith
m
e
ti
c
c
o
m
p
u
ti
n
g
:
A
g
e
n
e
ra
l
m
o
d
e
l
a
n
d
a
p
p
li
c
a
ti
o
n
s to
m
u
lt
i
p
li
e
r
a
n
d
sq
u
a
re
r
d
e
sig
n
,
”
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
si
u
m
o
n
L
o
w
P
o
we
r
El
e
c
tro
n
ics
a
n
d
De
sig
n
(I
S
L
PE
D),
v
o
l
.
6
4
,
n
o
.
4
,
p
p
.
1
0
8
1
-
1
0
9
0
,
2
0
1
4
.
[9
]
T.
A.
Dra
n
e
,
T
.
M
.
R
o
se
,
a
n
d
G
.
A.
C
o
n
sta
n
ti
n
id
e
s,
“
On
t
h
e
sy
s
tem
a
ti
c
c
re
a
ti
o
n
o
f
fa
it
h
f
u
ll
y
r
o
u
n
d
e
d
tru
n
c
a
ted
m
u
lt
ip
li
e
rs an
d
a
rra
y
s
,
”
IE
EE
T
r
a
n
sa
c
ti
o
n
s o
n
C
o
mp
u
ter
s,
v
o
l.
6
3
,
n
o
.
1
0
,
p
p
.
2
5
1
3
-
2
5
2
5
,
Oc
t.
2
0
1
4
.
[1
0
]
F
.
F
a
rsh
c
h
i,
M
.
S
.
Ab
r
ish
a
m
i,
S
.
M
.
F
a
k
h
ra
ie,
“
Ne
w
a
p
p
r
o
x
i
m
a
te
m
u
lt
ip
li
e
r
fo
r
l
o
w
p
o
we
r
d
i
g
it
a
l
sig
n
a
l
p
ro
c
e
ss
in
g
,
”
i
n
T
h
e
1
7
th
C
S
I
In
te
rn
a
ti
o
n
a
l
S
y
mp
o
siu
m
o
n
C
o
m
p
u
t
e
r
Arc
h
it
e
c
tu
re
&
Dig
it
a
l
S
y
ste
m
s
(CADS
2
0
1
3
)
,
2
0
1
3
,
p
p
.
2
5
-
30
.
[1
1
]
T.
Ya
m
a
m
o
to
,
I.
Tan
ig
u
c
h
i
,
H.
To
m
iy
a
m
a
,
S
.
Ya
m
a
sh
it
a
a
n
d
Y
.
Ha
ra
-
Az
u
m
i,
"
A
sy
ste
m
a
ti
c
m
e
th
o
d
o
l
o
g
y
fo
r
d
e
sig
n
a
n
d
wo
rst
-
c
a
se
e
rro
r
a
n
a
ly
sis
o
f
a
p
p
r
o
x
ima
te
a
rra
y
m
u
lt
ip
li
e
rs
,
"
IEI
CE
T
ra
n
s.
o
n
F
u
n
d
a
me
n
ta
ls,
v
o
l
.
1
0
0
,
n
o
.
7
,
p
p
.
1
4
9
6
-
1
4
9
9
,
2
0
1
7
.
[1
2
]
S
.
Bo
r
o
u
m
a
n
d
,
H.
P
.
Afs
h
a
r,
P
.
Brisk
a
n
d
S
.
M
o
h
a
m
m
a
d
i,
“
Ex
p
l
o
ra
ti
o
n
o
f
a
p
p
ro
x
ima
te
m
u
lt
ip
li
e
r
s
d
e
sig
n
sp
a
c
e
u
sin
g
c
a
rry
p
ro
p
a
g
a
ti
o
n
fre
e
c
o
m
p
re
ss
o
rs
,
”
in
2
0
1
8
2
3
r
d
Asia
a
n
d
S
o
u
th
P
a
c
if
ic
D
e
sig
n
A
u
to
m
a
ti
o
n
C
o
n
fer
e
n
c
e
(AS
P
-
DAC)
,
2
0
1
8
,
p
p
.
6
1
1
-
6
1
6
.
[1
3
]
C.
Li
u
,
J.
Ha
n
a
n
d
F
a
b
r
izio
Lo
m
b
a
rd
i,
“
A L
o
w
-
P
o
we
r,
Hig
h
-
p
e
rfo
r
m
a
n
c
e
a
p
p
ro
x
ima
te m
u
lt
ip
li
e
r
wit
h
c
o
n
fi
g
u
ra
b
le
p
a
rti
a
l
e
rro
r
re
c
o
v
e
ry
,
”
in
De
sig
n
,
Au
t
o
ma
ti
o
n
&
T
e
st i
n
Eu
ro
p
e
Co
n
fer
e
n
c
e
&
Exh
ib
it
i
o
n
(DAT
E)
,
2
0
1
4
,
p
p
.
1
-
4
.
[1
4
]
J.
Ch
a
n
g
a
n
d
J.
S
h
a
,
“
An
e
fficie
n
t
imp
lem
e
n
tatio
n
o
f
2
D
c
o
n
v
o
lu
ti
o
n
i
n
CNN
,
”
IEI
CE
El
e
c
tro
n
ics
Ex
p
re
ss
,
v
o
l.
1
4
,
n
o.
1,
p
p
.
1
-
8,
2
0
1
7
.
[1
5
]
B.
Li
u
,
H.
Qin
,
Y.
G
o
n
g
,
W.
G
e
,
M
.
Xia
a
n
d
L
.
S
h
i
,
”
EE
RA
-
ASR:
An
e
n
e
r
g
y
-
e
fficie
n
t
re
c
o
n
fi
g
u
ra
b
le
a
rc
h
it
e
c
tu
re
fo
r
a
u
to
m
a
ti
c
sp
e
e
c
h
re
c
o
g
n
it
io
n
with
h
y
b
ri
d
DN
N
a
n
d
a
p
p
ro
x
ima
te
c
o
m
p
u
ti
n
g
,
”
IEE
E
Acc
e
ss
,
v
o
l.
6
,
p
p
.
5
2
2
2
7
-
5
2
2
3
7
,
2
0
1
8
.
[1
6
]
D
-
A.
Ng
u
y
e
n
,
H
-
H.
H
o
,
D
-
H.
Bu
i
a
n
d
X
-
T
Tran
,
“
An
e
fficie
n
t
h
a
rd
wa
re
imp
lem
e
n
tati
o
n
o
f
a
rti
ficia
l
n
e
u
ra
l
n
e
two
rk
b
a
se
d
o
n
sto
c
h
a
stic
c
o
m
p
u
ti
n
g
,
”
in
NAF
O
S
T
ED
Co
n
f
e
re
n
c
e
o
n
In
fo
rm
a
ti
o
n
a
n
d
Co
mp
u
ter
S
c
ien
c
e
(NICS
)
,
2
0
1
8
,
p
p
.
2
3
7
-
2
4
2
.
[1
7
]
V.
M
ra
z
e
k
,
S
.
S
.
S
a
rwa
r,
L.
S
e
k
a
n
in
a
,
Z
.
Va
sic
e
k
a
n
d
K.
Ro
y
,
“
De
sig
n
o
f
p
o
we
r
-
e
fficie
n
t
a
p
p
ro
x
i
m
a
te
m
u
lt
ip
li
e
rs
fo
r
a
p
p
ro
x
ima
te
a
rti
ficia
l
n
e
u
ra
l
n
e
two
rk
s
,”
in
IEE
E/
AC
M
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
C
o
mp
u
ter
-
Ai
d
e
d
De
sig
n
(ICCAD)
,
2
0
1
6
,
p
p
.
1
-
7
.
[1
8
]
K.
S
imo
n
y
a
n
a
n
d
A.
Zi
ss
e
rm
a
n
,
”
Ve
ry
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
two
rk
s
fo
r
larg
e
-
sc
a
le
ima
g
e
r
e
c
o
g
n
it
i
o
n
,
”
in
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
L
e
a
rn
in
g
Rep
re
se
n
ta
ti
o
n
s,
2
0
1
4
.
[1
9
]
I.
Ha
m
m
a
d
a
n
d
K.
El
-
S
a
n
k
a
ry
,
“
I
m
p
a
c
t
o
f
a
p
p
ro
x
ima
te
m
u
lt
ip
li
e
rs
o
n
VG
G
d
e
e
p
lea
rn
in
g
n
e
two
r
k
,
”
IEE
E
Acc
e
ss
,
v
o
l.
6
,
p
p
.
6
0
4
3
8
-
6
0
4
4
4
,
2
0
1
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4
8
6
4
I
n
t J Reco
n
f
ig
u
r
a
b
le
&
E
m
b
ed
d
ed
Sy
s
t,
Vo
l.
10
,
No
.
1
,
Ma
r
c
h
2
0
2
1
:
1
–
10
10
[2
0
]
H.
S
im
a
n
d
J.
Lee
,
“
A
n
e
w
sto
c
h
a
stic
c
o
m
p
u
ti
n
g
m
u
lt
ip
li
e
r
with
a
p
p
li
c
a
ti
o
n
to
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
,”
in
IE
EE
De
sig
n
Au
to
ma
ti
o
n
Co
n
fer
e
n
c
e
(DAC),
2
0
1
7
,
p
p
.
1
-
6
.
[2
1
]
E.
S
c
h
k
u
fz
a
,
R.
S
h
a
rm
a
,
A.
Aik
e
n
,
"
S
t
o
c
h
a
stic
o
p
ti
m
iza
ti
o
n
o
f
f
lo
a
ti
n
g
-
p
o
in
t
p
r
o
g
ra
m
s
with
t
u
n
a
b
le
p
re
c
isio
n
"
,
ACM
S
IGP
L
AN
,
v
o
l.
4
9
,
n
o
.
6
,
p
p
.
5
3
-
6
4
,
2
0
1
4
.
[2
2
]
A.
Ala
g
h
i
a
n
d
J.
P
.
Ha
y
e
s,
“
S
u
rv
e
y
o
f
st
o
c
h
a
stic
c
o
m
p
u
ti
n
g
,
”
AC
M
T
r
a
n
s.
Em
b
e
d
d
e
d
C
o
mp
u
ti
n
g
S
y
ste
ms
,
v
o
l.
1
2
,
n
o
.
2
s,
p
p
.
1
-
1
9
,
2
0
1
2
.
[2
3
]
M
.
Co
u
rb
a
riau
x
a
n
d
Y.
Be
n
g
i
o
“
Bin
a
ry
Ne
t:
Trai
n
in
g
d
e
e
p
n
e
u
ra
l
n
e
two
r
k
s
with
we
i
g
h
ts
a
n
d
a
c
ti
v
a
ti
o
n
s
c
o
n
stra
i
n
e
d
to
+
1
o
r
−
1
,
”
a
rXiv
,
2
0
1
6
.
[2
4
]
N.
J.
F
ra
se
r,
Y.
Um
u
ro
g
lu
,
G
.
G
a
m
b
a
rd
e
ll
a
,
M
.
Blo
tt
,
P
.
Leo
n
g
,
M
.
Ja
h
re
a
n
d
K.
Viss
e
rs,
“
S
c
a
li
n
g
b
in
a
rize
d
n
e
u
ra
l
n
e
two
rk
s
o
n
re
c
o
n
fig
u
ra
b
le
l
o
g
ic
,
”
in
ACM
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
Pro
c
e
e
d
in
g
S
e
rie
s
(ACM
-
I
C
PS
)
,
2
0
1
7
,
p
p
.
25
-
30
.
[2
5
]
E.
Nu
rv
i
tad
h
i
,
D.
S
h
e
ffiel
d
,
J.
S
i
m
,
A.
M
ish
ra
,
G
.
Ve
n
k
a
tes
h
a
n
d
D.
M
a
rr,
“
a
c
c
e
ler
a
ti
n
g
b
i
n
a
rize
d
n
e
u
ra
l
n
e
two
r
k
s:
Co
m
p
a
riso
n
o
f
F
P
G
A,
CP
U,
G
P
U,
a
n
d
ASIC,
”
in
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
F
ield
-
Pro
g
ra
mm
a
b
le
T
e
c
h
n
o
l
o
g
y
(FP
T
)
,
2
0
1
6
,
p
p
.
77
-
84
.
[2
6
]
Y.
Zh
a
n
g
,
Z.
Zh
o
u
,
P
.
Hu
a
n
g
,
M
.
F
a
n
,
R.
Ha
n
,
W.
S
h
e
n
,
L
.
Li
u
,
X.
Li
u
a
n
d
J.
Ka
n
g
,
“
An
im
p
ro
v
e
d
h
a
rd
wa
re
a
c
c
e
lar
a
ti
o
n
a
rc
h
it
e
c
tu
re
o
f
b
i
n
a
ry
n
e
u
ra
l
n
e
two
r
k
wi
th
1
T1
R
a
rra
y
b
a
se
d
f
o
rwa
rd
/b
a
c
k
wa
rd
p
ro
p
a
g
a
ti
o
n
m
o
d
u
le
,
”
in
S
il
ico
n
Na
n
o
e
lec
tro
n
ics
W
o
rk
sh
o
p
(S
NW
),
2
0
1
9
,
p
p
.
1
-
2
.
[2
7
]
P.
K.
Ch
u
n
d
i,
P
.
Li
u
,
S
.
P
a
r
k
.
S
.
Lee
a
n
d
M
.
S
e
o
k
“
F
P
G
A
-
b
a
se
d
a
c
c
e
lera
ti
o
n
o
f
b
in
a
ry
n
e
u
ra
l
n
e
two
rk
trai
n
in
g
wit
h
m
in
imiz
e
d
o
ff
-
c
h
ip
m
e
m
o
ry
a
c
c
e
ss
,”
in
IEE
E/
ACM
I
n
ter
n
a
t
io
n
a
l
S
y
mp
o
si
u
m
o
n
L
o
w
P
o
we
r
El
e
c
tro
n
ics
a
n
d
De
sig
n
(IS
L
PE
D),
2
0
1
9
,
p
p
.
1
-
6
.
[2
8
]
Z.
Li
,
L.
Wan
g
,
S
.
G
u
o
,
Y.
De
n
g
,
Q.
Do
u
,
H
.
Z
h
o
u
a
n
d
W.
L
u
,
“
Laiu
s:
An
8
-
b
it
fi
x
e
d
-
p
o
i
n
t
CNN
h
a
rd
wa
re
in
fe
re
n
c
e
e
n
g
i
n
e
,
”
i
n
IE
EE
I
n
ter
n
a
ti
o
n
a
l
S
y
m
p
o
siu
m
o
n
Pa
ra
l
lel
a
n
d
Distrib
u
ted
Pr
o
c
e
ss
in
g
w
it
h
A
p
p
li
c
a
ti
o
n
s
a
n
d
2
0
1
7
I
EE
E
I
n
ter
n
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
Ub
i
q
u
i
to
u
s
Co
m
p
u
ti
n
g
a
n
d
C
o
mm
u
n
ic
a
ti
o
n
s
(I
S
PA
/IUCC)
,
2
0
1
7
,
p
p
.
1
4
3
-
150
.
[2
9
]
A.
Ba
laji,
S
.
Ullah
,
A.
Da
s,
a
n
d
A.
Ku
m
a
r
,
”
De
sig
n
m
e
th
o
d
o
l
o
g
y
f
o
r
e
m
b
e
d
d
e
d
a
p
p
ro
x
ima
te
a
rti
ficia
l
n
e
u
ra
l
n
e
two
rk
s,
”
i
n
Gr
e
a
t
L
a
k
e
s S
y
mp
o
siu
m o
n
VL
S
I
(GL
S
VL
S
I),
2
0
1
9
,
p
p
.
4
8
9
-
4
9
4
.
[3
0
]
A.
Kriz
h
e
v
sk
y
,
I
.
S
u
tsk
e
v
e
r
a
n
d
E.
Hin
t
o
n
,
“
Im
a
g
e
Ne
t
c
las
sifica
ti
o
n
wit
h
d
e
e
p
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
s
,
”
i
n
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ne
u
r
a
l
In
fo
rm
a
t
io
n
Pro
c
e
ss
in
g
S
y
ste
ms
,
v
o
l
.
6
0
,
n
o
.
6
,
p
p
.
8
4
-
9
0
,
2
0
1
7
.
B
I
O
G
RAP
H
I
E
S O
F
AUT
H
O
RS
Ke
n
ta
S
h
iran
e
re
c
e
iv
e
d
h
is
BE
d
e
g
re
e
in
e
lec
tro
n
ic
a
n
d
c
o
m
p
u
ter
e
n
g
i
n
e
e
rin
g
fr
o
m
Rit
su
m
e
ik
a
n
Un
iv
e
rsity
i
n
2
0
1
9
.
He
is
i
n
t
h
e
M
a
ste
r’s
d
e
g
re
e
p
ro
g
ra
m
a
t
Rit
su
m
e
ik
a
n
Un
i
v
e
rsity
.
His
re
se
a
rc
h
in
tere
sts in
c
lu
d
e
d
e
sig
n
m
e
th
o
d
o
lo
g
ies
fo
r
e
m
b
e
d
d
e
d
s
y
st
e
m
s.
Tak
a
h
iro
Ya
m
a
m
o
to
re
c
e
iv
e
d
h
i
s
BE
a
n
d
M
E
d
e
g
re
e
s
in
e
lec
tro
n
ic
a
n
d
c
o
m
p
u
ter
e
n
g
i
n
e
e
rin
g
fro
m
Rit
su
m
e
ik
a
n
Un
i
v
e
rsity
i
n
2
0
1
6
a
n
d
2
0
1
8
,
re
sp
e
c
ti
v
e
ly
.
At
p
re
se
n
t,
h
e
wo
r
k
s
f
o
r
M
it
su
b
ish
i
El
e
c
tri
c
Co
r
p
o
ra
t
io
n
.
His
re
se
a
rc
h
i
n
tere
sts
in
c
l
u
d
e
d
e
sig
n
m
e
th
o
d
o
lo
g
ie
s
fo
r
e
m
b
e
d
d
e
d
sy
ste
m
s.
Hiro
y
u
k
i
To
m
i
y
a
m
a
re
c
e
iv
e
d
h
is
BE,
M
E,
a
n
d
DE
d
e
g
re
e
s
i
n
c
o
m
p
u
ter
sc
ien
c
e
fro
m
K
y
u
sh
u
Un
iv
e
rsity
i
n
1
9
9
4
,
1
9
9
6
,
a
n
d
1
9
9
9
,
re
sp
e
c
ti
v
e
ly
.
He
wo
rk
e
d
a
s
a
v
isit
in
g
re
se
a
rc
h
e
r
a
t
UC
Irv
in
e
,
a
s
a
re
se
a
rc
h
e
r
a
t
IS
IT/
Ky
u
sh
u
,
a
n
d
a
s
a
n
a
ss
o
c
iate
p
ro
fe
ss
o
r
a
t
Na
g
o
y
a
U
n
iv
e
rsit
y
.
S
in
c
e
2
0
1
0
,
h
e
h
a
s
b
e
e
n
a
fu
ll
p
ro
fe
ss
o
r
wit
h
th
e
Co
ll
e
g
e
o
f
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
,
Rit
s
u
m
e
ik
a
n
Un
iv
e
rsity
.
He
h
a
s
se
rv
e
d
o
n
p
ro
g
ra
m
a
n
d
o
rg
a
n
izin
g
c
o
m
m
it
t
e
e
s
fo
r
a
n
u
m
b
e
r
o
f
p
re
m
ier
c
o
n
fe
re
n
c
e
s
in
c
lu
d
in
g
DA
C,
IC
CAD
,
DA
TE
,
ASP
-
DA
C,
COD
ES
+
IS
S
S
,
CASE
S
,
IS
LP
ED,
RTCS
A,
F
P
L,
a
n
d
M
P
S
o
C.
He
h
a
s
a
lso
se
rv
e
d
a
s
a
n
e
d
it
o
r
-
in
-
c
h
ief
fo
r
IP
S
J
TS
LDM
;
a
n
a
ss
o
c
iate
e
d
it
o
r
fo
r
ACM
TODA
ES
,
IEE
E
E
S
L,
a
n
d
S
p
ri
n
g
e
r
DA
EM
;
a
n
d
a
c
h
a
ir
fo
r
t
h
e
IEE
E
CS
Ka
n
sa
i
C
h
a
p
ter
a
n
d
IEE
E
C
EDA
Ja
p
a
n
C
h
a
p
ter.
His
re
se
a
rc
h
i
n
tere
sts
in
c
l
u
d
e
,
b
u
t
a
re
n
o
t
li
m
it
e
d
to
,
d
e
si
g
n
m
e
th
o
d
o
l
o
g
ies
fo
r
e
m
b
e
d
d
e
d
a
n
d
c
y
b
e
r
-
p
h
y
sic
a
l
sy
ste
m
s.
Evaluation Warning : The document was created with Spire.PDF for Python.