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.
1
6
,
No
.
4
,
A
u
g
u
s
t
20
2
6
,
p
p
.
1
964
~
1
9
7
5
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
4
.
pp
1
9
6
4
-
1
9
7
5
1964
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
A
hy
brid
retr
i
ev
a
l
a
ug
mente
d
g
ene
ra
tion
framewo
rk
for
a
utom
a
ted
educa
t
io
na
l docum
ent
u
ndersta
nding
and
int
ellig
en
t
respo
nse g
enerati
o
n
B
a
s
a
v
esh
D.
1
,
J
a
y
a
s
hree
Na
g
a
ra
j
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
J
y
o
t
h
y
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
B
a
n
g
a
l
o
r
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
-
A
I
M
L,
R
N
S
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
B
a
n
g
a
l
o
r
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
3
0
,
2
0
2
6
R
ev
is
ed
Ap
r
1
7
,
2
0
2
6
Acc
ep
ted
Ap
r
2
6
,
2
0
2
6
Ne
w
stu
d
e
n
ts
o
ften
stru
g
g
le
w
h
e
n
sh
o
rt
a
rti
c
les
c
las
h
wit
h
t
h
ick
t
e
x
tb
o
o
k
s.
S
ti
ll
,
e
v
e
n
th
o
u
g
h
larg
e
lan
g
u
a
g
e
m
o
d
e
ls
o
ffe
r
so
m
e
tea
c
h
in
g
su
p
p
o
rt
,
sta
n
d
a
rd
o
n
li
n
e
se
tu
p
s
lac
k
fo
c
u
s
e
d
a
c
c
u
ra
c
y
-
so
m
e
ti
m
e
s
m
a
k
in
g
th
in
g
s
up
-
a
n
d
risk
u
se
r
d
a
ta
c
o
n
tr
o
l.
He
re
c
o
m
e
s
a
n
id
e
a
:
b
u
il
d
a
ti
g
h
tl
y
te
ste
d
,
se
lf
-
c
o
n
tain
e
d
sy
ste
m
t
h
a
t
a
li
g
n
s
l
e
a
rn
in
g
m
a
teria
ls
a
u
to
m
a
ti
c
a
ll
y
with
o
u
t
n
e
e
d
in
g
t
h
e
i
n
tern
e
t,
k
e
e
p
i
n
g
in
f
o
rm
a
ti
o
n
p
riv
a
te
b
y
d
e
sig
n
.
On
e
lo
o
k
a
t
two
se
tu
p
s
sh
o
ws
h
o
w
t
h
e
y
h
a
n
d
le
l
o
c
a
l
re
a
so
n
in
g
d
iffere
n
tl
y
.
In
ste
a
d
o
f
u
sin
g
b
o
t
h
e
n
c
o
d
e
r
a
n
d
d
e
c
o
d
e
r
p
a
rts,
o
n
e
sy
ste
m
sk
i
p
s
t
h
e
e
n
c
o
d
e
r
e
n
t
irely
.
T
h
a
t
sim
p
ler
d
e
sig
n
g
ra
b
s
fu
l
l
c
o
n
te
x
t
th
ro
u
g
h
Ch
r
o
m
a
DB
with
o
u
t
sh
ri
n
k
i
n
g
t
h
e
d
a
ta
first.
M
e
a
n
wh
il
e
,
th
e
o
l
d
e
r
ty
p
e
c
ru
n
c
h
e
s
in
p
u
t
d
o
w
n
,
lo
si
n
g
m
e
a
n
in
g
a
lo
n
g
t
h
e
wa
y
.
Ev
e
n
t
h
o
u
g
h
it
ru
n
s
fa
st
-
ju
st
u
n
d
e
r
a
se
c
o
n
d
-
e
rro
rs
p
o
p
u
p
o
ften
,
f
o
u
r
o
u
t
o
f
fiv
e
re
sp
o
n
se
s
d
rift
in
g
o
ff
c
o
u
rse
.
On
th
e
fli
p
sid
e
,
th
e
n
e
w
m
e
th
o
d
b
u
il
d
s
c
o
rre
c
t
c
o
d
e
n
e
a
rly
e
v
e
ry
ti
m
e
,
a
d
d
s
c
lea
r
e
x
p
lan
a
ti
o
n
s
ti
e
d
t
o
les
so
n
g
o
a
ls,
y
e
t
tak
e
s
m
o
re
th
a
n
f
o
u
rtee
n
se
c
o
n
d
s
t
o
re
p
ly
.
S
lo
we
r?
Ye
s.
M
o
re
a
c
c
u
ra
te?
Clea
rly
.
Wh
a
t
sta
n
d
s
o
u
t
is
h
o
w
c
o
m
p
re
ss
e
d
m
o
d
e
l
s
ru
n
n
i
n
g
lo
c
a
ll
y
c
a
n
stil
l
c
a
tch
u
p
i
n
u
n
d
e
rsta
n
d
in
g
c
las
sro
o
m
c
o
n
ten
t
.
An
o
t
h
e
r
k
e
y
p
o
i
n
t
e
m
e
rg
e
s:
b
u
il
d
in
g
tu
t
o
rs
p
o
we
re
d
b
y
a
rti
ficia
l
in
tell
ig
e
n
c
e
(AI)
b
e
c
o
m
e
s sa
fe
r
wh
e
n
d
a
ta n
e
v
e
r
le
a
v
e
s th
e
d
e
v
ice
a
n
d
o
u
tsid
e
se
rv
ice
s
a
re
n
o
t
n
e
e
d
e
d
a
t
a
ll
.
K
ey
w
o
r
d
s
:
Au
to
m
ated
cu
r
r
icu
lu
m
m
ap
p
i
n
g
C
o
m
p
ar
ativ
e
ar
ch
itectu
r
e
an
aly
s
is
Data
s
o
v
er
eig
n
ty
in
ed
u
ca
tio
n
L
o
ca
l
lar
g
e
lan
g
u
ag
e
m
o
d
el
R
etr
iev
al
-
au
g
m
en
ted
g
en
er
ati
o
n
Vec
to
r
em
b
ed
d
in
g
s
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
:
J
ay
ash
r
ee
Nag
ar
aj
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
-
AI
ML
,
R
NS I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
B
an
g
alo
r
e,
I
n
d
ia
E
m
ail:
j
a
n
a
n
e
e
.
n
a
g
a
r
a
j@
g
m
a
il.
c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
ed
u
ca
tio
n
in
d
u
s
tr
y
is
in
cr
ea
s
in
g
ly
in
teg
r
atin
g
ar
tific
i
al
in
tellig
en
ce
(
AI
)
in
to
its
wo
r
k
in
g
s
to
ch
an
g
e
th
e
way
th
e
y
p
r
esen
t
o
r
co
n
s
u
m
e
lear
n
in
g
m
ater
ial.
Fo
r
co
m
p
u
ter
s
cien
ce
,
c
o
m
p
u
ter
to
o
ls
,
s
o
f
twar
e,
eq
u
ip
m
en
t,
a
n
d
/o
r
co
m
p
o
n
en
t
s
ar
e
u
s
ed
ex
ten
s
iv
ely
.
A
h
u
g
e
s
tu
d
y
m
ater
ial
is
to
b
e
co
v
er
ed
b
y
C
o
m
p
u
ter
Scien
ce
s
tu
d
en
ts
:
b
o
o
k
s
,
lectu
r
e
n
o
tes,
a
n
d
la
b
m
a
n
u
als.
Ho
wev
er
,
it
ca
n
b
e
q
u
ite
d
if
f
icu
lt
f
o
r
n
ew
s
tu
d
e
n
ts
to
m
an
ag
e
an
d
c
o
n
n
ec
t
s
u
ch
a
h
u
g
e
v
o
lu
m
e
of
s
tu
d
y
m
ate
r
ial.
T
h
e
s
y
llab
u
s
of
a
u
n
iv
e
r
s
ity
tells
wh
at
h
a
s
to
be
d
o
n
e.
B
u
t
n
ew
s
tu
d
en
ts
f
in
d
i
t
d
if
f
icu
lt
to
f
in
d
w
h
er
e
th
is
m
atter
is
ex
p
lain
ed
in
s
u
ch
la
r
g
e
v
o
lu
m
es.
Du
e
to
th
is
,
th
er
e
a
r
is
es
a
h
u
g
e
m
is
m
atch
b
etwe
en
wh
at
is
e
x
p
lain
e
d
in
th
e
u
n
iv
er
s
ity
s
y
llab
u
s
an
d
wh
at
is
p
r
o
v
id
e
d
in
s
tu
d
y
in
g
r
eso
u
r
ce
s
.
T
h
u
s
,
s
tu
d
en
ts
s
p
en
d
m
o
r
e
tim
e
tr
y
in
g
to
co
n
f
o
r
m
t
o
s
tu
d
y
m
ate
r
ial
p
atter
n
s
.
In
a
d
d
itio
n
to
th
eo
r
y
,
p
r
o
g
r
am
m
in
g
-
b
ased
la
b
wo
r
k
also
r
eq
u
ir
es
co
n
s
tan
t
g
u
id
a
n
ce
an
d
s
u
p
p
o
r
t,
a
n
d
lab
m
an
u
als,
alth
o
u
g
h
p
r
o
v
id
i
n
g
p
r
o
b
lem
ex
p
lan
atio
n
s
,
d
o
n
o
t
p
r
o
v
id
e
a
p
latf
o
r
m
to
v
er
i
f
y
an
s
wer
s
,
ass
ess
co
m
p
lex
ity
,
a
n
d
p
r
e
p
ar
e
f
o
r
v
iv
a
ex
am
in
atio
n
s
.
Stu
d
en
ts
o
f
ten
u
s
e
lar
g
e
lan
g
u
a
g
e
m
o
d
els
(
L
L
Ms)
,
s
u
ch
as
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
A
h
yb
r
id
r
etri
ev
a
l a
u
g
men
ted
g
en
era
tio
n
fr
a
mewo
r
k
fo
r
a
u
t
o
ma
ted
ed
u
ca
tio
n
a
l
…
(
B
a
s
a
v
esh
D.
)
1965
GPT
-
b
ased
m
o
d
els,
to
aid
th
e
m
in
s
u
ch
task
s
.
Alth
o
u
g
h
th
ese
m
o
d
els
ar
e
u
s
ef
u
l,
clo
u
d
-
b
ased
m
o
d
els
o
f
te
n
p
r
o
v
id
e
in
ac
cu
r
ate
a
n
d
am
b
ig
u
o
u
s
an
s
wer
s
,
wh
ich
d
o
n
o
t m
atch
ac
ad
em
ic
r
e
q
u
ir
em
en
ts
.
Nev
er
th
eless
,
th
e
two
cr
itical
is
s
u
es
r
elate
d
to
AI
tech
n
o
lo
g
y
’
s
s
wif
tly
g
r
o
win
g
ca
p
a
b
ilit
ies
ar
e
s
till
u
n
an
s
wer
ed
.
A
lim
itatio
n
o
f
L
L
Ms
is
th
at
th
ey
ca
n
n
o
t
ac
ce
s
s
in
s
titu
tio
n
-
s
p
ec
if
ic
lear
n
in
g
m
ater
ial,
m
ak
in
g
it
d
if
f
icu
lt
f
o
r
th
em
t
o
r
esp
o
n
d
a
s
p
er
p
r
escr
ib
ed
tex
tb
o
o
k
s
o
r
s
y
llab
u
s
es.
Yet
an
o
th
er
is
s
u
e
i
s
th
at
all
p
r
o
p
o
s
ed
s
o
lu
tio
n
s
tak
e
th
e
f
o
r
m
o
f
clo
u
d
-
b
ased
API
s
with
s
o
m
e
d
ata
p
r
iv
ac
y
an
d
laten
cy
is
s
u
es.
Po
s
tin
g
ed
u
ca
tio
n
al
d
o
cu
m
e
n
ts
o
n
,
s
ay
,
an
o
u
ts
id
e
s
er
v
er
m
a
y
n
o
t
b
e
f
ea
s
ib
le
in
a
s
ch
o
o
l
en
v
ir
o
n
m
e
n
t.
B
esid
es,
th
er
e
will
b
e
is
s
u
es with
s
er
v
er
lag
,
s
o
d
ea
lin
g
with
s
u
ch
s
y
s
tem
s
m
ay
tak
e
tim
e.
T
h
is
wo
r
k
aim
s
to
p
r
o
p
o
s
e
a
s
y
s
tem
f
o
r
s
o
lv
in
g
th
e
is
s
u
es
o
f
d
ata
p
r
iv
ac
y
a
n
d
ac
ad
em
ic
c
u
r
r
icu
lu
m
alig
n
m
en
t
with
o
u
t
r
ely
i
n
g
o
n
ex
ter
n
al
s
er
v
er
s
.
I
t
is
b
ased
o
n
a
r
etr
ie
v
al
-
au
g
m
e
n
ted
g
en
er
atio
n
(
R
AG)
f
r
am
ewo
r
k
,
wh
er
e
to
p
ics
in
a
s
y
llab
u
s
ca
n
b
e
lin
k
e
d
with
r
ef
er
en
ce
m
ater
ials
with
o
u
t
r
ely
in
g
o
n
ex
ter
n
a
l
s
er
v
er
s
.
I
t
in
clu
d
es
a
p
ar
s
in
g
m
ec
h
an
is
m
f
o
r
ex
t
r
ac
tin
g
s
tr
u
ctu
r
ed
d
ata
f
r
o
m
u
n
o
r
g
an
ize
d
PDF
d
o
cu
m
en
ts
.
Mo
r
eo
v
er
,
th
is
wo
r
k
in
clu
d
es
two
d
if
f
er
en
t
ap
p
r
o
ac
h
es
f
o
r
s
o
lv
in
g
th
is
p
r
o
b
lem
:
o
n
e
is
b
ased
o
n
a
b
aselin
e
ap
p
r
o
ac
h
u
s
in
g
an
en
co
d
e
r
-
d
ec
o
d
er
m
o
d
el
with
co
m
p
r
ess
ed
r
etr
iev
al,
an
d
an
o
th
e
r
ap
p
r
o
ac
h
is
b
ased
on
a
d
ec
o
d
er
-
o
n
ly
m
o
d
el
u
s
in
g
f
u
ll
-
co
n
tex
t r
etr
ie
v
al
f
o
r
ac
cu
r
ac
y
.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
of
th
i
s
wo
r
k
ar
e
th
r
ee
f
o
ld
.
First,
we
p
r
o
p
o
s
e
a
co
m
p
a
r
ativ
e
ev
al
u
atio
n
of
v
ar
io
u
s
ar
ch
itectu
r
es
f
o
r
l
o
ca
l
R
AG
s
y
s
tem
s
,
with
a
f
o
cu
s
o
n
ac
h
iev
in
g
a
tr
ad
e
-
o
f
f
b
etwe
en
s
p
ee
d
an
d
ac
cu
r
ac
y
.
Seco
n
d
,
we
p
r
o
p
o
s
e
a
m
o
d
u
lar
d
u
al
p
ip
elin
e
s
tr
u
ctu
r
e,
s
p
ec
if
ically
d
esig
n
ed
f
o
r
ed
u
ca
tio
n
al
s
ce
n
ar
io
s
in
v
o
lv
in
g
th
e
o
r
y
a
n
d
p
r
ac
tice.
T
h
ir
d
,
we
s
h
o
w
h
o
w
co
m
p
lex
R
AG
-
b
ased
wo
r
k
f
lo
ws
ca
n
b
e
p
er
f
o
r
m
ed
o
n
co
n
s
u
m
e
r
-
g
r
a
d
e
h
ar
d
war
e
with
q
u
an
tized
m
o
d
els,
allo
win
g
f
o
r
f
u
lly
o
f
f
lin
e
d
ep
lo
y
m
en
t
with
o
u
t sacr
if
icin
g
d
ata
p
r
iv
ac
y
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
Ar
tific
ial
I
n
tellig
en
ce
in
ed
u
ca
tio
n
al
tech
h
as
co
m
e
a
lo
n
g
way
.
It
s
tar
ted
with
b
asic,
r
u
le
-
b
ased
s
y
s
tem
s
,
an
d
n
o
w
we’
v
e
g
o
t
m
o
d
els
th
at
ca
n
ac
tu
ally
u
n
d
e
r
s
tan
d
co
n
te
x
t
an
d
g
e
n
er
ate
n
u
an
ce
d
r
esp
o
n
s
es.
I
n
th
is
r
ev
iew,
th
e
au
th
o
r
s
d
ig
in
to
h
o
w
th
ese
tech
n
o
lo
g
ies
h
av
e
p
r
o
g
r
ess
ed
,
esp
ec
ially
lo
o
k
in
g
at
wh
er
e
L
L
Ms
f
all
s
h
o
r
t
lik
e
wh
en
y
o
u
p
u
t
t
h
em
to
wo
r
k
in
s
tr
ict
ac
ad
em
i
c
s
ettin
g
s
.
T
h
e
au
th
o
r
s
also
ex
p
lo
r
e
R
AG,
wh
ich
s
tan
d
s
o
u
t
as
a
p
o
wer
f
u
l
way
to
im
p
r
o
v
e
co
n
tex
t
u
al
ac
cu
r
ac
y
.
T
h
e
r
ev
iew
g
ets
in
to
th
e
n
u
ts
an
d
b
o
lts
,
co
m
p
ar
in
g
en
c
o
d
er
-
d
ec
o
d
er
with
d
ec
o
d
e
r
-
o
n
ly
m
o
d
els,
es
p
ec
ially
f
o
r
s
itu
atio
n
s
wh
e
r
e
h
ar
d
war
e
is
lim
ited
an
d
in
f
er
en
ce
n
ee
d
s
to
h
ap
p
e
n
lo
ca
lly
.
L
o
o
k
in
g
th
r
o
u
g
h
r
e
ce
n
t
s
tu
d
ies,
o
n
e
th
in
g
is
clea
r
:
m
o
s
t
clo
u
d
-
b
ased
R
AG
f
r
am
ewo
r
k
s
d
o
a
g
r
ea
t
jo
b
with
g
e
n
er
al
q
u
esti
o
n
a
n
d
an
s
wer
task
s
.
B
u
t
th
er
e
i
s
a
r
ea
l
g
ap
w
h
en
it
co
m
es to
s
tr
o
n
g
,
f
u
lly
lo
ca
l sy
s
tem
s
b
u
ilt f
o
r
h
an
d
lin
g
co
m
p
l
ex
,
m
u
lti
-
m
o
d
al
ed
u
ca
tio
n
al
d
o
cu
m
en
ts
an
d
o
n
es
th
at
ac
tu
ally
k
ee
p
d
ata
s
ec
u
r
e
an
d
s
o
v
er
eig
n
.
T
h
er
e
i
s
ju
s
t
n
o
t
m
u
ch
o
u
t
th
er
e
th
at
m
ee
ts
th
ese
to
u
g
h
r
eq
u
ir
em
e
n
ts
[
1
]
.
2
.
1
.
G
ener
a
t
iv
e
AI
a
nd
t
he
h
a
llu
cina
t
io
n
c
h
a
l
l
en
g
e
Sy
s
tem
s
b
u
ilt
o
n
T
r
a
n
s
f
o
r
m
er
d
esig
n
s
[
2
]
,
u
tili
zin
g
o
p
tim
iz
ed
m
ath
em
atica
l
ac
tiv
atio
n
f
u
n
ctio
n
s
to
p
r
o
ce
s
s
d
ee
p
lin
g
u
is
tic
h
ier
ar
c
h
ies
[
3
]
,
s
h
o
w
h
o
w
lar
g
e
lan
g
u
ag
e
m
o
d
els
m
ig
h
t
ac
t
lik
e
tu
t
o
r
in
g
to
o
ls
.
B
r
o
wn
an
d
team
f
o
u
n
d
th
ese
m
o
d
els
lear
n
f
ast
f
r
o
m
lim
ited
ex
am
p
l
es
-
th
an
k
s
to
tr
ain
i
n
g
o
n
m
ass
iv
e
tex
t
co
llectio
n
s
-
s
o
th
e
y
a
n
s
wer
q
u
esti
o
n
s
o
r
ex
p
lain
id
ea
s
with
o
u
t
n
ee
d
in
g
ex
tr
a
s
etu
p
f
o
r
ea
ch
task
[
4
]
.
Sti
ll,
u
s
in
g
th
em
in
class
r
o
o
m
s
f
ac
es
a
b
ig
h
u
r
d
le:
s
o
m
etim
es
th
ey
in
v
en
t
an
s
wer
s
th
at
s
o
u
n
d
r
ig
h
t
b
u
t
ar
e
n
o
t
tr
u
e,
esp
ec
ially
wh
en
ask
ed
a
b
o
u
t
t
o
p
ics
m
i
s
s
in
g
f
r
o
m
th
eir
o
r
ig
i
n
al
d
at
a
[
5
]
.
T
h
at
g
litch
b
ec
o
m
es
a
s
er
io
u
s
is
s
u
e
at
u
n
iv
er
s
ities
,
wh
er
e
e
v
er
y
d
e
f
in
itio
n
m
u
s
t
m
atch
th
e
tex
t
b
o
o
k
ex
ac
tly
.
B
ec
au
s
e
o
f
th
is
,
g
en
er
al
-
p
u
r
p
o
s
e
m
o
d
els
ca
n
n
o
t
wo
r
k
al
o
n
e
as
r
eliab
le
h
elp
er
s
in
th
o
s
e
s
ettin
g
s
,
r
eq
u
ir
in
g
r
o
b
u
s
t
m
ec
h
a
n
is
m
s
to
r
estrict
th
eir
g
en
er
atio
n
s
co
p
e
to
v
er
if
ied
c
o
r
p
u
s
es
[
6
]
,
[
7
]
.
2
.
2
.
Ret
rie
v
a
l
-
a
ug
m
ent
ed
g
e
nera
t
io
n
W
h
en
m
o
d
els
ca
n
n
o
t
lear
n
n
ew
f
ac
ts
af
ter
tr
ain
in
g
,
r
esear
ch
er
s
f
o
u
n
d
a
wo
r
k
a
r
o
u
n
d
.
I
n
s
tead
o
f
r
ely
in
g
o
n
ly
o
n
wh
at
th
e
m
o
d
el
m
em
o
r
ized
,
t
h
ey
b
e
g
an
p
u
l
lin
g
in
o
u
ts
id
e
in
f
o
r
m
atio
n
d
u
r
in
g
an
s
wer
s
.
On
e
m
eth
o
d
u
s
es
s
to
r
ed
tex
ts
th
at
g
et
s
ea
r
ch
ed
wh
en
n
ee
d
ed
.
T
h
e
s
y
s
tem
g
r
ab
s
p
iece
s
o
f
th
o
s
e
tex
ts
r
ig
h
t
b
ef
o
r
e
r
esp
o
n
d
in
g
[
8
]
,
f
o
r
m
alizin
g
an
en
d
-
to
-
en
d
f
r
am
ewo
r
k
wh
er
e
p
ar
am
etr
ic
m
em
o
r
y
is
r
e
in
f
o
r
ce
d
with
n
o
n
-
p
ar
am
etr
ic
ex
ter
n
al
k
n
o
wled
g
e
[
9
]
.
T
h
at
way
,
r
esp
o
n
s
es c
o
m
e
f
r
o
m
u
p
-
to
-
d
ate
s
o
u
r
ce
s
,
n
o
t ju
s
t o
ld
d
ata
s
tu
ck
in
th
e
m
o
d
el
[
8
]
,
a
p
r
in
ci
p
le
th
at
h
o
ld
s
tr
u
e
wh
et
h
er
r
etr
i
ev
in
g
co
n
tex
t
at
in
f
er
e
n
ce
ti
m
e
o
r
wh
en
s
ca
lin
g
r
etr
iev
al
ar
ch
itectu
r
es
ac
r
o
s
s
tr
illi
o
n
s
o
f
p
r
e
-
tr
ain
in
g
to
k
e
n
s
[
1
0
]
.
I
t
is
lik
e
g
iv
in
g
th
e
m
o
d
el
ac
ce
s
s
to
n
o
tes
wh
ile
an
s
wer
in
g
q
u
esti
o
n
s
.
Fa
cts
s
tay
s
h
ar
p
er
b
ec
au
s
e
th
ey
’
r
e
p
u
lled
f
r
esh
ea
ch
tim
e.
At
f
ir
s
t,
th
ese
s
y
s
tem
s
u
s
ed
b
asic r
etr
iev
al
s
tep
s
.
Ov
e
r
tim
e,
s
m
ar
ter
way
s
o
f
p
ick
in
g
an
d
u
s
in
g
in
f
o
s
tar
ted
ap
p
ea
r
in
g
T
h
an
k
s
to
ad
v
a
n
ce
s
lik
e
Sen
ten
ce
-
B
E
R
T
f
r
o
m
R
eim
er
s
an
d
Gu
r
ev
y
c
h
[
1
1
]
,
d
e
n
s
e
v
ec
to
r
s
n
o
w
r
u
n
with
o
u
t
clo
u
d
s
.
R
u
n
n
in
g
lo
ca
lly
m
ea
n
s
s
ea
r
ch
es
k
ee
p
m
ea
n
in
g
in
tact
wh
ile
s
tay
in
g
o
f
f
l
in
e.
Priv
ac
y
g
ain
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
1
96
4
-
1
975
1966
s
tr
en
g
th
wh
en
d
ata
n
e
v
er
lea
v
es
th
e
d
ev
ice.
Har
d
war
e
o
n
ce
to
o
wea
k
n
o
w
h
a
n
d
les
wh
at
n
ee
d
ed
s
er
v
e
r
s
b
ef
o
r
e.
T
h
ese
s
h
if
ts
o
p
en
d
o
o
r
s
q
u
ietly
,
with
o
u
t f
a
n
f
ar
e.
2
.
3
.
Arc
hite
ct
ura
l
c
o
ns
idera
t
io
ns
:
E
nco
der
-
D
ec
o
der
v
s
.
D
ec
o
der
-
o
n
l
y
Pick
in
g
an
in
f
e
r
en
ce
en
g
in
e
m
atter
s
a
lo
t
wh
en
b
u
ild
in
g
lo
ca
l
R
AG
s
etu
p
s
-
it
s
h
ap
es
h
o
w
f
ast
r
esp
o
n
s
es
co
m
e
v
er
s
u
s
h
o
w
d
ee
p
th
e
th
in
k
i
n
g
ca
n
g
o
.
T
5
m
o
d
els,
p
ar
t
o
f
th
e
E
n
co
d
e
r
-
Dec
o
d
er
f
am
ily
,
g
et
u
s
ed
a
lo
t
b
ec
au
s
e
th
e
y
h
an
d
l
e
s
u
m
m
ar
y
wo
r
k
a
n
d
lan
g
u
ag
e
s
wap
s
with
o
u
t
s
lo
win
g
th
i
n
g
s
d
o
wn
[
1
2
]
.
Sti
ll,
n
ee
d
in
g
cu
s
to
m
tr
ain
i
n
g
f
o
r
ea
ch
jo
b
m
ea
n
s
th
ey
f
ail
to
ad
ap
t
wh
en
f
ac
ed
with
m
ess
y
,
lay
er
ed
p
r
o
b
lem
s
[
1
3
]
.
W
h
en
f
ed
ch
u
n
k
s
p
u
lled
f
r
o
m
lo
n
g
s
tr
etch
es
o
f
ac
ad
em
ic
te
x
ts
,
th
ese
m
o
d
els
ten
d
to
s
tall
r
ath
er
th
an
co
n
n
ec
t
d
is
tan
t id
ea
s
s
m
o
o
th
ly
[
1
4
]
.
Yet
th
in
g
s
ch
an
g
ed
wh
en
d
e
co
d
er
-
o
n
ly
s
etu
p
s
b
e
g
an
h
a
n
d
lin
g
lo
ca
l
task
s
d
if
f
er
e
n
tly
.
Mistra
l
7
B
ca
m
e
o
u
t,
b
u
ilt
b
y
J
ian
g
et
a
l.
[
1
5
]
,
s
h
o
win
g
co
m
p
ac
t
d
esig
n
s
ca
n
m
atch
b
u
l
k
ier
s
y
s
tem
s
lik
e
L
L
aM
A
1
3
B
o
n
to
u
g
h
lo
g
ic
an
d
co
d
e
jo
b
s
[
1
6
]
.
B
ec
au
s
e
o
f
tr
ick
s
lik
e
s
tep
-
by
-
s
tep
p
r
o
m
p
ts
-
W
ei
et
a
l.
in
tr
o
d
u
ce
d
th
em
[
1
7
]
.
T
h
ese
m
o
d
els
n
o
w
d
o
b
etter
i
n
g
r
a
d
in
g
c
o
d
ed
wo
r
k
au
to
m
a
tically
[
1
8
]
,
[
1
9
]
.
I
n
s
tead
o
f
s
k
ip
p
in
g
ah
ea
d
,
th
is
p
ap
er
lin
es
u
p
o
n
e
s
tan
d
ar
d
e
n
co
d
er
-
d
ec
o
d
er
m
eth
o
d
b
esid
e
a
n
ew
d
ec
o
d
er
-
o
n
ly
s
etu
p
,
tied
to
g
eth
e
r
u
s
in
g
to
o
ls
ca
lled
L
an
g
C
h
ain
[
2
0
]
a
n
d
C
h
r
o
m
aDB
[
2
1
]
,
ch
ec
k
in
g
h
o
w
ea
c
h
clo
s
es
th
e
s
p
ac
e
b
e
twee
n
co
u
r
s
e
p
lan
s
an
d
tex
tb
o
o
k
s
.
3.
M
E
T
H
O
DO
L
O
G
Y
B
u
ilt
lik
e
p
iece
s
o
f
a
p
u
zz
le,
t
h
is
s
etu
p
r
u
n
s
f
ir
s
t
o
n
y
o
u
r
d
e
v
ice
u
s
in
g
s
ep
ar
ate
p
ar
ts
jo
in
e
d
to
g
et
h
er
-
R
ea
ct
h
an
d
les
wh
at
y
o
u
s
ee
,
Flas
k
m
an
ag
es
b
eh
in
d
-
t
h
e
-
s
ce
n
es
task
s
.
R
u
n
n
in
g
th
e
b
r
ai
n
s
is
a
R
AG
m
eth
o
d
,
g
u
id
ed
b
y
L
an
g
C
h
ain
[
2
0
]
,
li
n
k
in
g
d
ata
r
etr
iev
al
with
an
s
wer
g
en
er
atio
n
.
T
h
r
ee
m
ain
s
ec
tio
n
s
s
h
ap
e
h
o
w
it
wo
r
k
s
:
o
n
e
p
r
ep
a
r
es
in
co
m
in
g
in
f
o
r
m
atio
n
,
an
o
th
e
r
s
to
r
es
m
ea
n
in
g
-
r
ich
k
n
o
wled
g
e,
w
h
ile
th
e
th
ir
d
b
u
ild
s
r
esp
o
n
s
es wh
en
ask
ed
s
o
m
eth
i
n
g
.
No
t
ev
er
y
s
etu
p
wo
r
k
s
th
e
s
a
m
e
u
n
d
er
tig
h
t
h
ar
d
war
e
lim
its
,
s
o
test
in
g
b
ec
am
e
ess
en
tial.
On
e
p
ath
s
tu
ck
to
s
h
r
in
k
in
g
v
ec
to
r
s
,
cu
ttin
g
d
o
wn
p
r
o
ce
s
s
in
g
d
em
an
d
s
-
th
is
s
er
v
ed
as
th
e
s
tar
tin
g
p
o
in
t.
An
o
th
er
tr
ie
d
k
ee
p
in
g
all
co
n
tex
t
in
tact,
aim
in
g
f
o
r
r
ich
e
r
m
ea
n
in
g
in
r
esu
lts
.
T
h
r
ee
s
tag
es
s
h
ap
ed
h
o
w
th
in
g
s
u
n
f
o
ld
e
d
:
s
h
ap
in
g
d
esig
n
s
s
id
e
b
y
s
id
e,
s
ettin
g
u
p
r
ea
l
-
wo
r
ld
test
s
,
th
en
ch
o
o
s
in
g
w
h
at
to
m
ea
s
u
r
e
a
f
t
er
war
d
.
3
.
1
.
Co
m
pa
ra
t
iv
e
a
rc
hite
ct
ura
l
d
e
s
i
g
n
A
p
air
o
f
test
s
etu
p
s
to
o
k
s
h
a
p
e,
aim
ed
at
weig
h
i
n
g
h
o
w
in
f
er
en
ce
s
p
ee
d
m
atch
es
u
p
wit
h
lear
n
in
g
s
tr
u
ctu
r
e.
On
e
co
n
s
id
er
ed
tim
in
g
f
ir
s
t,
wh
ile
th
e
o
th
er
lean
ed
in
to
c
o
u
r
s
e
co
h
e
r
en
ce
.
E
ac
h
p
lay
ed
out
u
n
d
e
r
s
lig
h
tly
s
h
if
ted
co
n
d
itio
n
s
,
r
ev
ea
lin
g
wh
er
e
d
ela
y
s
b
u
m
p
ed
a
g
ain
s
t in
s
tr
u
ctio
n
al
f
lo
w
.
a.
Star
tin
g
with
m
o
d
el
A:
T
h
e
b
aselin
e
u
s
in
g
co
m
p
r
ess
ed
r
etr
i
ev
al
-
it
r
elied
o
n
th
e
c
o
m
p
ac
t
Flan
-
T
5
s
etu
p
to
s
et
a
lo
wer
b
o
u
n
d
in
p
er
f
o
r
m
an
ce
.
B
ec
au
s
e
lim
ited
co
n
tex
t sp
ac
e
p
o
s
ed
ch
allen
g
es,
an
ap
p
r
o
ac
h
s
im
ilar
to
R
E
FR
AG
wa
s
tes
ted
[
2
2
]
,
wh
er
e
n
ar
r
ativ
e
s
eg
m
en
ts
p
u
lled
d
u
r
in
g
r
et
r
iev
al
wer
e
tu
r
n
ed
in
to
d
en
s
e
v
ec
to
r
s
u
m
m
a
r
ies
p
r
io
r
to
g
e
n
er
atio
n
.
Yet
r
esu
lts
f
r
o
m
r
e
m
o
v
in
g
co
m
p
o
n
en
ts
s
h
o
we
d
m
ea
n
in
g
o
f
ten
f
ad
ed
wh
e
n
o
n
l
y
th
ese
co
n
d
en
s
ed
v
er
s
io
n
s
f
ed
th
e
g
en
er
ato
r
,
u
n
lik
e
f
ee
d
in
g
c
o
m
p
lete
p
ass
ag
es d
ir
ec
tly
.
b.
A
f
r
esh
tak
e
o
n
m
o
d
el
B
em
er
g
es:
T
h
is
o
n
e
lean
f
u
lly
o
n
r
aw
co
n
tex
t.
B
u
ilt
ar
o
u
n
d
a
q
u
an
tized
Mistra
l
7
B
s
etu
p
,
it
s
k
ip
s
an
y
f
o
r
m
o
f
d
ata
s
h
r
in
k
i
n
g
.
I
n
s
id
e
liv
es
a
C
h
r
o
m
aDB
m
o
d
u
le
[
2
1
]
,
q
u
ietly
h
o
ld
i
n
g
u
n
alter
ed
c
h
u
n
k
s
o
f
lear
n
in
g
m
ater
ial
in
d
ex
e
d
u
s
in
g
h
ier
ar
ch
ical
n
av
ig
a
b
le
s
m
all
wo
r
l
d
(
HNSW
)
g
r
ap
h
m
atch
in
g
[
2
3
]
.
T
h
ese
p
iece
s
f
lo
w
s
tr
aig
h
t
in
to
th
e
m
o
d
el
’
s
8
k
-
to
k
e
n
s
p
ac
e,
n
o
f
ilter
s
ap
p
l
ied
.
B
y
f
ee
d
in
g
u
n
to
u
c
h
ed
p
ass
ag
es,
it
tap
s
d
ee
p
ly
in
to
h
o
w
d
ec
o
d
e
r
-
o
n
ly
m
o
d
els
lear
n
f
r
o
m
im
m
e
d
iate
s
u
r
r
o
u
n
d
in
g
s
.
Acc
u
r
ac
y
r
is
es wh
en
th
e
s
y
s
te
m
s
ee
s
ev
er
y
th
in
g
,
ex
ac
tly
as
-
is
.
3
.
2
.
P
ha
s
e
1:
H
euristic
s
t
ruct
ura
l
e
x
t
r
a
c
t
i
o
n
R
ig
h
t
af
ter
g
r
ab
b
i
n
g
s
y
llab
i
or
lab
m
a
n
u
als,
th
ey
g
et
clea
n
ed
up
in
to
a
u
n
if
o
r
m
lay
o
u
t
f
ir
s
t.
Ou
r
p
ar
s
in
g
to
o
l,
b
u
ilt
with
R
eg
E
x
p
atter
n
s
,
tack
les
m
ess
y
PDF
lay
o
u
ts
h
ea
d
-
o
n
b
y
f
ilter
in
g
o
u
t
clu
tter
au
to
m
atica
lly
.
a.
B
r
ea
k
in
g
d
o
wn
th
e
s
y
llab
u
s
h
ap
p
en
s
wh
en
th
e
s
y
s
tem
s
p
o
ts
s
ec
tio
n
s
lab
eled
lik
e
Mo
d
u
le
-
1
o
r
s
im
ilar
.
W
h
er
e
n
u
m
b
er
s
f
o
llo
w
"M
o
d
u
le"
it g
r
ab
s
th
o
s
e
ch
u
n
k
s
f
o
r
p
r
o
ce
s
s
in
g
.
T
ex
t th
at
ju
s
t states
tim
e
len
g
th
s
o
r
b
o
o
k
n
a
m
es g
ets lef
t b
eh
in
d
.
W
h
at
r
em
ain
s
is
a
s
tr
aig
h
tf
o
r
war
d
s
et
o
f
s
u
b
jects r
ea
d
y
f
o
r
s
ea
r
ch
.
E
ac
h
item
in
th
is
co
llectio
n
-
ca
ll
th
e
m
₁
th
r
o
u
g
h
ₙ
-
is
p
u
lled
d
ir
ec
tly
f
r
o
m
wh
at
i
s
v
is
ib
le
o
n
th
e
p
ag
e.
Ma
tch
in
g
wo
r
k
s
ev
en
if
letter
s
ar
e
m
ix
ed
u
p
p
er
o
r
lo
wer
ca
s
e.
R
aw
te
x
t f
r
o
m
PDFs
f
ee
d
s
in
to
th
is
s
t
ep
with
o
u
t p
r
io
r
clea
n
u
p
.
b.
W
h
en
wo
r
k
in
g
with
lab
g
u
id
es,
th
e
m
eth
o
d
u
s
es
a
f
lex
ib
le
ed
g
e
-
f
in
d
i
n
g
p
r
o
ce
s
s
.
Scan
n
in
g
b
eg
in
s
at
m
ar
k
er
s
lik
e
"L
is
t
o
f
ex
p
er
im
en
ts
"
to
s
p
o
t
en
tr
y
p
o
in
ts
.
Gr
o
u
p
in
g
s
m
aller
task
s
-
s
u
ch
as
1
a
,
1
b
-
h
ap
p
en
s
n
ex
t,
f
o
r
m
in
g
u
n
if
ie
d
u
n
its
f
o
r
q
u
er
ies.
L
o
g
ical
f
lo
w
s
tay
s
i
n
tact
th
r
o
u
g
h
later
s
tag
es
o
f
a
u
to
m
atic
s
cr
ip
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
A
h
yb
r
id
r
etri
ev
a
l a
u
g
men
ted
g
en
era
tio
n
fr
a
mewo
r
k
fo
r
a
u
t
o
ma
ted
ed
u
ca
tio
n
a
l
…
(
B
a
s
a
v
esh
D.
)
1967
cr
ea
tio
n
.
c.
W
h
en
th
e
s
o
f
twar
e
s
p
o
ts
b
o
t
h
u
n
i
q
u
e
p
atter
n
s
to
g
eth
er
,
it
m
ar
k
s
th
e
f
ile
as
a
f
u
ll
h
y
b
r
id
co
u
r
s
e
wh
ile
tu
r
n
in
g
o
n
an
in
ter
ac
tiv
e
m
o
d
e
f
o
r
m
ix
e
d
task
s
.
T
h
at
s
etu
p
g
iv
es
lear
n
er
s
t
h
e
ch
a
n
ce
t
o
s
witch
b
ac
k
an
d
f
o
r
th
-
s
tu
d
y
i
n
g
less
o
n
g
o
als
o
n
e
m
o
m
en
t,
t
h
en
w
o
r
k
in
g
th
r
o
u
g
h
h
a
n
d
s
-
o
n
p
r
o
g
r
a
m
m
in
g
ch
allen
g
es
r
i
g
h
t
af
ter
-
all
in
s
id
e
th
e
s
am
e
s
tr
et
ch
o
f
tim
e.
T
h
e
co
m
p
lete
s
em
an
tic
in
d
ex
i
n
g
wo
r
k
f
lo
w,
d
etailin
g
th
e
t
r
a
n
s
f
o
r
m
atio
n
o
f
u
n
s
tr
u
ctu
r
e
d
P
DF
tex
t
in
to
v
ec
to
r
em
b
ed
d
in
g
s
,
is
d
ep
icted
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
T
h
e
Sem
an
tic
I
n
d
ex
i
n
g
W
o
r
k
f
lo
w
d
etailin
g
th
e
p
r
o
ce
s
s
of
tr
an
s
f
o
r
m
in
g
u
n
s
tr
u
ctu
r
ed
PDF
tex
t
in
to
v
ec
to
r
em
b
e
d
d
in
g
s
f
o
r
s
to
r
a
g
e
in
C
h
r
o
m
aDB
3
.
3
.
P
ha
s
e
2:
Sem
a
ntic
ind
ex
ing
a
nd
r
e
t
r
i
e
v
a
l
C
lo
s
e
at
h
an
d
,
an
s
wer
s
to
o
k
s
h
ap
e
f
r
o
m
s
to
r
ed
less
o
n
s
.
Nea
r
b
y
s
h
elv
es
f
ed
th
e
s
y
s
te
m
'
s
t
h
o
u
g
h
t
s
.
L
ec
tu
r
es
an
d
r
ea
d
in
g
lis
ts
b
ec
am
e
its
f
o
u
n
d
atio
n
.
T
ex
tb
o
o
k
s
f
o
r
m
ed
a
co
r
e
o
t
h
er
s
f
o
llo
w
ed
.
Pro
x
im
ity
g
a
v
e
weig
h
t to
wh
at
it k
n
ew.
C
am
p
u
s
life
an
ch
o
r
e
d
ev
e
r
y
r
e
p
ly
.
a.
Star
tin
g
with
th
e
tex
tb
o
o
k
,
it
g
ets
s
p
lit
b
y
a
p
r
o
ce
s
s
th
at
b
r
ea
k
s
tex
t
s
tep
b
y
s
tep
.
A
b
it
p
ass
es
in
to
th
e
f
o
llo
win
g
o
n
e,
ed
g
es
to
u
ch
in
g
,
ea
ch
p
ar
t
lar
g
e
en
o
u
g
h
to
k
e
ep
v
ital
in
f
o
in
tact.
Me
a
n
in
g
s
tay
s
tr
u
e
with
in
ev
er
y
c
u
t
y
et
f
l
o
ws
n
atu
r
all
y
f
r
o
m
ea
r
lier
lin
es.
T
in
y
s
ec
tio
n
s
p
r
eser
v
e
e
x
ac
t
ex
p
la
n
atio
n
s
.
E
ar
lier
ex
p
lan
atio
n
s
s
et
th
e
s
ce
n
e,
wh
ich
m
ea
n
s
th
in
g
s
d
o
n
o
t
s
ee
m
d
is
co
n
n
ec
ted
.
B
ec
au
s
e
o
f
th
is
f
lo
w,
em
b
ed
d
in
g
s
r
ef
lect
h
o
w
t
h
o
u
g
h
ts
r
ea
lly
d
ev
elo
p
in
th
e
o
r
ig
in
al
tex
t.
b.
First
u
p
,
p
iece
s
o
f
wr
itten
m
e
an
in
g
b
ec
am
e
tig
h
t
n
u
m
b
er
c
o
d
es
b
y
way
o
f
th
e
Hu
g
g
in
g
Fa
ce
all
-
Min
iLM
-
L6
-
v
2
s
etu
p
[
1
1
]
.
W
o
r
d
s
wer
e
n
o
t
h
an
d
led
o
n
e
at
a
tim
e
-
in
s
tead
,
en
tire
th
o
u
g
h
ts
f
o
ld
e
d
in
to
s
tr
in
g
s
o
f
3
8
4
n
u
m
b
er
s
.
T
h
e
ap
p
r
o
ac
h
s
k
ip
s
f
u
ll
ex
p
lan
atio
n
s
y
et
h
o
l
d
s
co
n
ce
p
ts
in
a
f
o
r
m
at
m
ath
ca
n
wo
r
k
with
.
No
t
ju
s
t
m
atch
in
g
k
e
y
wo
r
d
s
,
it
lean
s
o
n
h
a
b
its
th
e
T
r
an
s
f
o
r
m
er
p
ick
ed
u
p
wh
ile
lear
n
in
g
.
Un
d
er
n
ea
th
,
ev
er
y
s
n
ip
p
et
f
in
d
s
its
s
p
o
t in
a
s
p
ac
e
with
m
an
y
d
im
en
s
io
n
s
,
wh
er
e
lik
e
m
ea
n
in
g
s
clu
s
ter
n
ea
r
ea
ch
o
th
er
.
c.
E
ac
h
tim
e
a
lear
n
er
p
o
s
es
a
q
u
esti
o
n
,
th
e
s
y
s
tem
ch
ec
k
s
h
o
w
m
u
c
h
it
r
esem
b
les
s
to
r
e
d
p
ass
ag
es
u
s
in
g
co
s
in
e
s
im
ilar
ity
.
R
ath
er
th
an
m
atch
in
g
wo
r
d
s
,
it
lo
o
k
s
at
u
n
d
er
ly
i
n
g
id
ea
s
th
r
o
u
g
h
n
u
m
er
ical
v
ec
to
r
s
.
W
h
en
co
m
p
ar
in
g
th
e
in
c
o
m
i
n
g
q
u
er
y
to
k
n
o
wn
d
ata
p
o
in
ts
,
p
r
o
x
im
ity
d
eter
m
in
es
f
it.
C
lo
s
est
m
atch
es
g
u
id
e
wh
ich
m
ater
ials
co
m
e
f
o
r
war
d
.
Usu
ally
,
th
r
ee
s
eg
m
en
ts
s
u
r
f
ac
e
b
ased
o
n
alig
n
m
en
t.
Me
an
in
g
s
h
ap
es
r
etr
iev
al
m
o
r
e
t
h
an
e
x
ac
t
p
h
r
asin
g
d
o
es.
On
ly
h
ig
h
est
-
s
co
r
in
g
f
r
ag
m
e
n
ts
en
ter
th
e
n
ex
t
s
tag
e
.
Fo
cu
s
s
tay
s
n
ar
r
o
w
-
ju
s
t
th
o
s
e
s
elec
ted
b
its
m
o
v
e
ah
ea
d
.
B
ac
k
wh
at
r
etu
r
n
s
wo
r
k
s
r
ig
h
t w
h
er
e
it
b
elo
n
g
s
.
Ma
th
with
v
ec
to
r
s
ch
ec
k
s
th
at
o
n
e.
3
.
4
.
P
ha
s
e
3:
P
ro
m
pt
eng
ineering
a
nd
g
e
n
e
r
a
t
i
o
n
Du
r
in
g
in
f
er
e
n
ce
,
a
f
ix
ed
s
ch
em
a
g
u
id
es
th
e
p
r
o
m
p
ts
s
o
o
u
tp
u
ts
s
tay
u
n
if
o
r
m
en
o
u
g
h
f
o
r
g
r
ad
in
g
.
W
h
at
co
m
es
o
u
t
f
its
wh
at
'
s
ex
p
ec
ted
wh
en
s
co
r
in
g
wo
r
k
.
Stru
ctu
r
e
k
ee
p
s
th
in
g
s
p
r
ed
ictab
l
e
with
o
u
t
s
u
r
p
r
is
es
in
f
o
r
m
at.
E
ac
h
r
esp
o
n
s
e
lin
es
up
b
ec
au
s
e
r
u
les
s
h
ap
e
h
o
w
q
u
esti
o
n
s
ar
e
a
s
k
ed
.
Pre
d
ictab
ilit
y
m
atter
s
m
o
s
t
o
n
ce
an
s
wer
s
s
tar
t f
lo
win
g
.
Sc
h
em
a
lo
ck
s
d
o
wn
ch
o
ices e
ar
ly
o
n
p
u
r
p
o
s
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
1
96
4
-
1
975
1968
A
d
if
f
er
e
n
t
k
in
d
o
f
th
in
k
in
g
k
ick
s
in
wh
en
t
h
e
s
y
s
tem
o
n
ly
u
s
es
p
r
o
v
id
e
d
tex
tb
o
o
k
p
iece
s
.
B
y
lo
ck
in
g
an
s
wer
s
to
ju
s
t
th
o
s
e
s
ec
tio
n
s
,
m
ad
e
ce
n
tr
al
r
ig
h
t
f
r
o
m
th
e
s
tar
t,
wild
g
u
ess
es
f
ad
e
o
u
t.
Mo
s
t
b
ig
m
o
d
els
ten
d
to
in
v
en
t
d
etails
if
lef
t
u
n
ch
ec
k
ed
.
T
h
at
co
m
m
o
n
f
law?
I
t
s
h
r
in
k
s
h
er
e
b
ec
au
s
e
im
ag
in
atio
n
g
ets
tied
d
o
wn
to
wr
itten
f
ac
ts
p
u
ll
ed
s
tr
aig
h
t f
r
o
m
s
o
u
r
ce
m
ater
i
al.
Star
ti
n
g
o
f
f
,
t
h
e
r
e'
s
a
s
et
u
p
ca
l
led
t
h
e
"T
ea
ch
in
g
ass
is
ta
n
t
"
r
o
le
u
s
ed
d
u
r
i
n
g
c
o
d
i
n
g
t
as
k
s
.
T
h
is
f
o
r
m
at
ask
s
f
o
r
r
es
p
o
n
s
es
s
p
lit
i
n
t
o
c
l
ea
r
s
ec
ti
o
n
s
u
s
i
n
g
Ma
r
k
d
o
w
n
.
O
n
e
p
a
r
t
m
u
s
t
d
esc
r
i
b
e
h
o
w
t
h
e
s
o
l
u
ti
o
n
w
o
r
k
s
,
s
tep
b
y
s
t
ep
,
i
n
p
la
in
wo
r
d
s
.
F
o
ll
o
w
in
g
t
h
a
t
c
o
m
es
t
h
e
ac
t
u
a
l
w
o
r
k
i
n
g
c
o
d
e,
wr
i
tte
n
p
r
o
p
e
r
ly
i
n
e
it
h
e
r
P
y
t
h
o
n
o
r
C
.
T
h
en
l
ast
,
an
es
tim
at
e
s
h
o
ws
h
o
w
m
u
ch
ti
m
e
an
d
m
em
o
r
y
it
u
s
es
,
ex
p
l
ai
n
e
d
wit
h
B
ig
-
O
n
o
t
ati
o
n
.
Star
tin
g
elsewh
er
e
ea
c
h
tim
e
,
f
ac
ts
g
et
p
u
lled
f
ir
s
t,
th
en
s
en
t
en
ce
s
b
u
ilt
af
ter
.
T
h
is
way
s
p
l
its
f
in
d
in
g
in
f
o
f
r
o
m
m
ak
in
g
an
s
wer
s
.
O
n
e
p
ar
t
g
r
a
b
s
d
ata
lik
e
an
o
ld
-
s
ch
o
o
l
s
ea
r
ch
er
wo
u
ld
d
o
.
T
h
e
o
th
er
s
h
ap
es
wo
r
d
s
s
m
o
o
th
ly
,
m
u
ch
lik
e
to
d
ay
'
s
talk
-
lik
e
s
y
s
tem
s
b
eh
av
e.
Sep
a
r
atin
g
th
ese
jo
b
s
h
elp
s
m
ix
p
r
ec
is
io
n
with
n
atu
r
al
f
lo
w.
Do
n
e
r
i
g
h
t,
tr
u
th
s
tay
s
in
tact
wh
ile
p
h
r
asin
g
s
o
u
n
d
s
le
s
s
s
tiff
.
4.
I
M
P
L
E
M
E
NT
A
T
I
O
N
A
ND
E
XP
E
R
I
M
E
N
T
A
L
SE
T
UP
A
f
r
esh
s
etu
p
ca
m
e
t
o
g
eth
er
as
a
s
elf
-
co
n
tain
ed
we
b
to
o
l
m
ea
n
t
to
r
u
n
with
o
u
t
in
ter
n
et,
s
tay
in
g
in
s
id
e
p
r
iv
ate
n
etwo
r
k
s
o
n
ly
.
B
u
ilt
o
n
f
r
ee
ly
av
ailab
le
t
o
o
ls
s
o
an
y
o
n
e
ca
n
r
e
b
u
ild
it,
k
ee
p
co
s
ts
lo
w,
s
tay
in
co
n
tr
o
l
o
f
in
f
o
r
m
atio
n
.
I
ts
b
ac
k
b
o
n
e
r
u
n
s
Py
th
o
n
3
.
1
0
h
a
n
d
lin
g
s
er
v
er
d
u
ties
wh
ile
R
ea
ct
s
h
ap
es
h
o
w
u
s
er
s
in
ter
ac
t.
T
h
ese
p
ar
ts
talk
th
r
o
u
g
h
s
tan
d
ar
d
API
lin
k
s
m
ad
e
f
o
r
s
m
o
o
th
ex
ch
an
g
e.
4
.
1
.
Dev
el
o
pm
ent
env
iro
n
ment
a
nd
d
e
p
en
de
n
ci
e
s
B
u
ilt
o
n
a
Fla
s
k
f
o
u
n
d
atio
n
,
th
e
s
y
s
tem
h
an
d
les
r
eq
u
ests
q
u
ietly
b
eh
in
d
th
e
s
ce
n
es.
L
ig
h
tweig
h
t
b
y
d
esig
n
,
it
wo
r
k
s
s
m
o
o
t
h
ly
with
to
o
ls
co
d
e
d
in
Py
th
o
n
f
o
r
s
m
ar
t
p
r
o
ce
s
s
in
g
task
s
.
T
o
lin
k
p
ar
ts
s
af
ely
,
s
h
ar
ed
r
eso
u
r
ce
s
talk
ac
r
o
s
s
d
o
m
ain
s
u
s
in
g
an
ad
d
e
d
C
OR
S
h
elp
er
.
C
o
m
m
u
n
icatio
n
s
tay
s
tig
h
t
a
n
d
tr
u
s
ted
b
etwe
en
u
s
er
in
ter
f
ac
e
an
d
lo
ca
l m
o
d
el
r
u
n
n
e
r
.
Mo
s
t
of
th
e
t
o
o
ls
u
s
ed
h
er
e
ar
e
b
u
ilt
on
th
ese
co
d
e
c
o
l
l
e
c
ti
o
n
s
:
a.
A
web
o
f
c
o
n
n
ec
tio
n
s
b
etwe
en
to
o
ls
-
lo
a
d
er
s
,
s
to
r
a
g
e,
b
r
ain
-
r
u
n
s
o
n
L
an
g
C
h
ain
[
2
0
]
.
T
h
is
s
y
s
tem
k
ee
p
s
ea
ch
p
iece
talk
i
n
g
at
th
e
r
ig
h
t
tim
e.
I
ts
j
o
b
is
m
a
k
in
g
s
u
r
e
s
tep
s
f
o
llo
w
o
n
e
an
o
th
er
with
o
u
t
h
iccu
p
s
.
Pieces sh
if
t o
n
ly
wh
en
o
th
e
r
s
f
in
is
h
th
eir
wo
r
k
.
b.
Fro
m
tim
e
to
tim
e,
lab
s
taf
f
p
u
ll d
ata
o
u
t o
f
PDFs
u
s
in
g
a
to
o
l c
alled
Py
Mu
PDF,
k
n
o
wn
in
s
id
e
th
e
co
d
e
as
f
itz
.
I
ts
s
p
ee
d
s
tan
d
s
o
u
t
wh
en
g
r
ab
b
i
n
g
tex
t
f
ast
with
o
u
t
lo
s
in
g
s
tr
u
ctu
r
e.
Un
lik
e
s
lo
wer
to
o
ls
s
u
ch
as
Py
PDF2
,
it
k
ee
p
s
f
o
r
m
attin
g
in
tact
m
o
r
e
r
eliab
ly
.
W
h
en
m
an
u
als
h
av
e
o
d
d
lay
o
u
ts
o
r
m
ix
ed
f
o
n
ts
,
th
is
o
n
e
h
an
d
les
th
em
b
etter
.
Per
f
o
r
m
an
ce
m
atter
s
m
o
s
t
h
er
e
-
esp
ec
ially
o
n
d
e
n
s
e
p
ag
es
f
illed
with
ch
ar
ts
an
d
f
o
o
tn
o
tes.
Oth
er
o
p
tio
n
s
wer
e
test
ed
,
y
et
th
ey
lag
g
e
d
b
eh
i
n
d
d
u
r
in
g
b
u
lk
p
r
o
ce
s
s
in
g
.
Pre
cisi
o
n
in
s
p
ac
in
g
an
d
lin
e
o
r
d
er
m
ak
es
th
e
d
if
f
er
en
ce
n
o
ticea
b
le.
So
,
ev
en
if
less
co
m
m
o
n
,
t
h
is
p
ick
f
it
s
th
e
wo
r
k
f
lo
w
tig
h
ter
.
c.
R
u
n
n
in
g
in
s
id
e
th
e
a
p
p
,
C
h
r
o
m
aDB
[
2
1
]
ac
ts
as
a
b
u
ilt
-
in
m
em
o
r
y
b
a
n
k
f
o
r
v
ec
to
r
s
.
W
ith
o
u
t
n
ee
d
in
g
o
u
ts
id
e
s
er
v
er
s
lik
e
Pin
ec
o
n
e
or
Milv
u
s
,
it
k
ee
p
s
ev
er
y
th
in
g
s
elf
-
co
n
tain
ed
.
Desig
n
ed
t
h
is
way
,
it
f
it
s
p
er
f
ec
tly
with
in
s
tr
ict
o
f
f
lin
e
b
o
u
n
d
ar
ies
-
n
o
in
ter
n
et,
n
o
ex
ce
p
tio
n
s
.
T
h
e
w
h
o
le
s
y
s
te
m
s
tay
s
lo
ca
l
b
y
d
ef
au
lt.
T
h
e
h
ig
h
-
le
v
el
s
y
s
tem
ar
ch
itectu
r
e,
illu
s
tr
atin
g
t
h
e
i
n
ter
ac
tio
n
b
etwe
en
th
e
f
r
o
n
ten
d
,
b
ac
k
en
d
,
an
d
in
f
er
en
ce
e
n
g
in
e,
is
s
h
o
wn
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Hig
h
-
lev
el
s
y
s
tem
ar
ch
itectu
r
e
of
t
h
e
lo
ca
l
R
AG
ap
p
licatio
n
illu
s
tr
atin
g
th
e
in
ter
a
ctio
n
b
etwe
en
th
e
R
ea
ct
f
r
o
n
ten
d
Flas
k
b
ac
k
en
d
an
d
th
e
lo
ca
l i
n
f
er
en
ce
en
g
in
e
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
A
h
yb
r
id
r
etri
ev
a
l a
u
g
men
ted
g
en
era
tio
n
fr
a
mewo
r
k
fo
r
a
u
t
o
ma
ted
ed
u
ca
tio
n
a
l
…
(
B
a
s
a
v
esh
D.
)
1969
4
.
2
.
L
o
ca
l
LLM
c
o
nfig
ura
t
io
n
(
O
l
l
a
ma
)
R
u
n
n
in
g
o
n
y
o
u
r
o
wn
m
ac
h
i
n
e,
th
e
s
y
s
tem
u
s
es
Ollam
a
in
s
t
ea
d
o
f
s
en
d
in
g
d
ata
t
h
r
o
u
g
h
clo
u
d
API
s
.
T
h
at
k
ee
p
s
p
e
r
s
o
n
al
in
f
o
r
m
ati
o
n
o
u
t
o
f
r
em
o
te
s
er
v
e
r
s
b
y
p
r
o
ce
s
s
in
g
it
lo
ca
lly
.
I
n
s
tead
o
f
r
ely
in
g
o
n
o
u
ts
id
e
co
n
n
ec
tio
n
s
,
e
v
er
y
th
i
n
g
h
a
p
p
e
n
s
wh
er
e
y
o
u
co
n
t
r
o
l it.
a.
Star
tin
g
with
m
o
d
el
p
ic
k
-
Mistra
l
7
B
g
o
t
th
e
n
o
d
,
p
u
ll
ed
f
r
o
m
a
lin
eu
p
o
f
o
p
tio
n
s
.
I
ts
ed
g
e?
Stro
n
g
r
esu
lts
o
n
lo
g
ic
an
d
p
r
o
g
r
a
m
m
in
g
ch
allen
g
es,
esp
ec
ially
c
o
n
s
id
er
in
g
h
o
w
lig
h
tweig
h
t
it
is
.
Per
f
o
r
m
an
ce
p
er
p
ar
a
m
eter
tip
p
ed
th
e
s
ca
le
h
er
e.
b.
T
o
o
p
e
r
ate
th
is
7
-
b
illi
o
n
-
p
ar
a
m
eter
m
o
d
el
o
n
r
eg
u
la
r
lap
to
p
s
u
s
in
g
o
n
ly
8
GB
to
1
6
GB
o
f
m
em
o
r
y
,
we
ap
p
lied
4
-
b
it
q
u
an
tizatio
n
-
s
p
ec
if
ically
QL
o
R
A,
a
m
eth
o
d
o
u
tlin
ed
b
y
Dettm
er
s
an
d
tea
m
[
2
4
]
.
B
ec
au
s
e
of
th
is
ap
p
r
o
ac
h
,
th
e
r
eq
u
ir
ed
C
P
U
an
d
VR
AM
u
s
ag
e
d
r
o
p
s
s
h
ar
p
ly
,
f
allin
g
f
r
o
m
ar
o
u
n
d
2
8
g
i
g
ab
y
tes
d
o
wn
to
ju
s
t
o
v
er
4
.
1
.
As
a
r
esu
lt,
s
tu
d
en
ts
can
n
o
w
r
u
n
it
ev
en
wh
en
th
ey
lack
s
p
ec
ialized
GPU
eq
u
ip
m
en
t.
c.
A
b
r
id
g
e
f
o
r
m
s
wh
en
Py
th
o
n
talk
s
to
Ollam
a
th
r
o
u
g
h
L
an
g
C
h
ain
-
O
llam
a,
lettin
g
p
r
o
m
p
t
s
f
lo
w
with
o
u
t
d
elay
.
T
h
is
p
ath
k
ee
p
s
d
ata
m
o
v
in
g
s
tr
aig
h
t
in
to
th
e
l
o
ca
l
m
o
d
el.
Stre
am
in
g
s
tay
s
s
m
o
o
th
b
ec
au
s
e
th
e
lin
k
cu
ts
ex
tr
a
s
tep
s
.
I
n
s
tead
o
f
wai
tin
g
,
r
esp
o
n
s
es st
ar
t f
ast.
T
h
e
s
etu
p
r
u
n
s
q
u
ietly
b
eh
in
d
ev
e
r
y
r
eq
u
est.
4
.
3
.
Vec
t
o
r
em
bedd
ing
i
mp
l
e
me
n
t
a
t
i
o
n
W
o
r
d
s
g
et
tu
r
n
ed
in
to
n
u
m
b
e
r
p
atter
n
s
th
at
h
elp
m
ac
h
in
es
u
n
d
er
s
tan
d
m
ea
n
in
g
.
T
h
ese
p
atter
n
s
let
s
y
s
tem
s
s
p
o
t
s
im
ilar
ities
b
ey
o
n
d
ex
ac
t
w
o
r
d
m
atch
es.
Un
d
er
s
tan
d
in
g
g
r
o
ws
wh
e
n
p
h
r
ases
m
ap
clo
s
e
to
g
et
h
er
in
a
h
id
d
en
s
p
ac
e.
Ma
tch
in
g
s
h
if
ts
f
r
o
m
k
ey
wo
r
d
s
to
h
o
w
id
ea
s
r
elate
.
Me
an
in
g
b
ec
o
m
es
s
o
m
eth
in
g
m
ath
ca
n
m
e
a
s
u
r
e
.
a.
A
s
m
all
T
r
an
s
f
o
r
m
er
s
y
s
tem
tu
r
n
s
s
en
ten
ce
s
in
to
f
ix
ed
-
s
iz
e
n
u
m
b
er
b
u
n
d
les.
B
u
ilt
o
n
Sen
ten
ce
-
B
E
R
T
id
ea
s
,
it
u
s
es
th
e
all
-
Min
iLM
-
L6
-
v
2
s
etu
p
f
r
o
m
Hu
g
g
i
n
g
Fa
ce
[
1
1
]
.
E
ac
h
p
h
r
ase
b
ec
o
m
es
a
p
o
in
t
in
3
8
4
d
im
en
s
io
n
s
th
r
o
u
g
h
t
h
is
m
ap
p
in
g
p
r
o
ce
s
s
.
b.
B
ec
au
s
e
it
h
an
d
les
s
en
ten
ce
m
atch
in
g
well,
th
is
m
o
d
el
r
u
n
s
s
m
o
o
th
ly
ev
en
o
n
r
e
g
u
lar
c
o
m
p
u
te
r
p
r
o
ce
s
s
o
r
s
.
W
ith
o
u
t
n
ee
d
in
g
h
ea
v
y
h
ar
d
war
e
,
p
r
o
ce
s
s
in
g
d
o
cu
m
en
ts
s
tay
s
f
ast
an
d
s
m
o
o
th
.
T
h
e
d
esig
n
s
k
ip
s
th
e
u
s
u
al
lag
th
at
co
m
es
f
r
o
m
r
ely
in
g
o
n
ad
v
an
ce
d
g
r
ap
h
ics
ch
ip
s
.
E
f
f
icien
c
y
win
s
wh
en
ev
e
r
y
d
a
y
m
ac
h
in
es c
an
k
ee
p
u
p
with
o
u
t
s
p
ec
ial
g
ea
r
.
A
v
is
u
aliza
tio
n
in
d
icatin
g
h
o
w
th
e
all
-
Min
iLM
-
L6
-
v
2
m
o
d
el
h
an
d
les
th
e
s
em
an
tic
clu
s
ter
in
g
o
f
r
elate
d
tex
tb
o
o
k
co
n
ce
p
ts
is
p
r
esen
ted
in
Fig
u
r
e
3
.
Fig
u
r
e
3
.
Vis
u
aliza
tio
n
of
v
ec
to
r
s
p
ac
e
em
b
e
d
d
in
g
s
d
em
o
n
s
t
r
atin
g
th
e
s
em
an
tic
cl
u
s
ter
in
g
of
r
elate
d
tex
tb
o
o
k
co
n
ce
p
ts
u
s
in
g
th
e
all
-
Min
iLM
-
L6
-
v
2
m
o
d
el
4
.
4
.
Appl
ica
t
io
n
wo
r
k
f
lo
w
i
mp
l
e
me
n
t
a
t
i
o
n
On
e
p
ath
d
ea
l
with
s
tu
d
en
t
r
e
co
r
d
s
s
ep
ar
ately
.
An
o
t
h
er
h
a
n
d
les
en
r
o
lm
en
t
s
tep
s
th
r
o
u
g
h
a
d
if
f
er
en
t
ch
an
n
el.
E
ac
h
f
u
n
ctio
n
co
n
n
e
cts
v
ia
its
o
wn
ad
d
r
ess
o
n
lin
e.
Sp
ec
if
ic
ac
tio
n
s
lin
k
to
in
d
i
v
id
u
al
ad
d
r
ess
es
s
et
asid
e
ju
s
t f
o
r
th
em
a.
Star
tin
g
o
f
f
,
th
e
s
y
s
tem
tak
e
s
in
f
iles
u
s
in
g
a
m
eth
o
d
th
at
s
p
lits
u
p
lo
ad
s
in
to
ch
u
n
k
s
f
o
r
s
m
o
o
th
e
r
h
an
d
lin
g
.
T
h
en
co
m
es
id
e
n
tific
atio
n
-
p
atter
n
s
s
p
o
t
w
h
eth
er
a
d
o
c
u
m
en
t
is
a
PDF
t
h
r
o
u
g
h
tar
g
ete
d
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
1
96
4
-
1
975
1970
m
atch
in
g
r
u
les.
T
h
at
ea
r
ly
ch
ec
k
m
ak
es
s
u
r
e
o
n
ly
n
ea
t,
o
r
g
an
ized
d
ata
m
o
v
es
f
o
r
war
d
.
I
f
s
k
ip
p
ed
,
m
ess
y
lay
o
u
t
b
its
s
lip
th
r
o
u
g
h
.
T
h
o
s
e
g
litch
es
clu
tter
m
ea
n
i
n
g
late
r
o
n
.
R
esp
o
n
s
es
g
et
wo
r
s
e
wh
en
b
ac
k
g
r
o
u
n
d
n
o
is
e
s
n
ea
k
s
in
to
an
aly
s
is
.
b.
Ou
t
of
th
e
b
lu
e,
it
p
u
lls
m
ater
ial
f
r
o
m
th
e
s
y
llab
u
s
_
ch
r
o
m
a_
d
b
to
s
h
ap
e
th
eo
r
y
d
etails.
Su
d
d
en
ly
,
th
e
s
y
s
tem
tr
ig
g
er
s
th
e
lo
ca
l
Mistra
l
L
L
M
with
a
tem
p
er
atu
r
e
s
e
t
at
0
.
3
.
B
ec
au
s
e
p
r
ec
is
io
n
m
a
tter
s
,
r
esp
o
n
s
es
s
tay
clo
s
e
to
f
ac
ts
f
o
u
n
d
i
n
tex
tb
o
o
k
p
iece
s
.
No
t
lo
n
g
af
ter
r
etr
ie
v
al,
ex
p
la
n
atio
n
s
f
o
r
m
th
r
o
u
g
h
co
n
tr
o
lled
,
r
ep
ea
tab
le
lo
g
ic.
E
v
en
s
o
,
r
an
d
o
m
n
ess
s
tay
s
l
o
w
-
ev
er
y
o
u
tp
u
t
s
tick
s
n
ea
r
k
n
o
wn
c
o
n
ten
t.
W
h
ile
p
r
o
ce
s
s
in
g
,
co
n
tex
t
lin
k
s
tig
h
tly
to
wh
at
th
e
d
atab
ase
p
r
o
v
id
es.
On
ly
th
en
d
o
es
s
tr
u
ctu
r
ed
ex
p
lan
atio
n
e
m
er
g
e
f
r
o
m
alig
n
ed
f
r
a
g
m
en
ts
.
c.
W
h
atev
er
co
m
es
n
e
x
t
d
ep
e
n
d
s
o
n
th
e
la
b
an
s
wer
s
b
ein
g
g
en
er
ated
f
ir
s
t.
T
h
is
s
tep
lin
es
u
p
ev
er
y
th
in
g
n
ee
d
ed
f
o
r
th
e
la
b
p
i
p
elin
e
to
r
u
n
s
m
o
o
th
l
y
.
A
u
n
iq
u
e
p
r
o
m
p
t
tem
p
late
ta
k
es
ch
ar
g
e
h
e
r
e,
s
h
ap
in
g
h
o
w
th
in
g
s
tu
r
n
o
u
t.
T
h
e
f
o
r
m
at
m
u
s
t
s
tick
to
Ma
r
k
d
o
wn
r
u
l
es
-
n
o
e
x
ce
p
tio
n
s
.
C
o
d
e
ap
p
ea
r
s
f
ir
s
t,
t
h
en
co
m
p
lex
ity
lev
el,
f
o
llo
wed
b
y
v
iv
a
q
u
esti
o
n
s
.
E
ac
h
p
iece
s
lo
ts
in
to
p
lace
ex
ac
tly
wh
er
e
it
s
h
o
u
ld
.
R
ep
o
r
ts
co
m
e
o
u
t
u
n
if
o
r
m
b
ec
a
u
s
e
o
f
t
h
is
s
etu
p
.
Stru
ctu
r
e
g
u
id
es e
v
e
r
y
o
u
tc
o
m
e
b
e
h
in
d
th
e
s
ce
n
es.
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
o
ev
alu
ate
th
e
ef
f
icac
y
o
f
th
e
p
r
o
p
o
s
ed
o
f
f
lin
e
au
to
m
ated
cu
r
r
icu
lu
m
alig
n
m
e
n
t
s
y
s
tem
,
we
co
n
d
u
cte
d
a
s
er
ies o
f
q
u
alitativ
e
an
d
q
u
an
titativ
e
test
s
u
s
in
g
s
tan
d
ar
d
u
n
d
er
g
r
ad
u
ate
c
o
m
p
u
ter
s
cien
ce
s
y
llab
i
an
d
co
r
r
esp
o
n
d
i
n
g
lab
o
r
ato
r
y
m
an
u
als.
T
h
e
s
y
s
tem
was
ass
es
s
ed
on
th
r
ee
p
r
im
ar
y
m
etr
ics:
C
o
n
tex
tu
al
ac
cu
r
ac
y
(
v
e
r
if
y
in
g
g
r
o
u
n
d
i
n
g
in
th
e
p
r
o
v
id
ed
r
ef
e
r
en
ce
tex
tb
o
o
k
)
,
Sy
n
tactica
l
co
r
r
ec
tn
ess
(
test
in
g
ex
ec
u
tab
ilit
y
o
f
A
I
-
g
en
e
r
ated
co
d
e)
,
a
n
d
s
y
s
tem
in
f
er
e
n
ce
la
ten
cy
(
m
ea
s
u
r
in
g
g
e
n
er
atio
n
r
esp
o
n
s
e
tim
es).
5
.
1
.
Q
ua
lit
a
t
iv
e
a
na
ly
s
is
:
Sy
lla
bu
s
-
t
ex
t
bo
o
k
a
l
i
g
n
me
n
t
Star
tin
g
o
f
f
,
th
e
m
ain
g
o
al
b
e
h
in
d
t
h
e
"T
h
e
o
r
y
p
ip
elin
e
"
w
as
m
ak
in
g
s
u
r
e
e
v
er
y
ex
p
la
n
a
tio
n
s
tu
ck
clo
s
ely
to
th
e
g
iv
en
s
o
u
r
ce
.
W
h
en
we
tr
ied
it
o
u
t,
u
s
in
g
s
o
m
eth
in
g
lik
e
"Po
ly
m
o
r
p
h
is
m
in
Py
th
o
n
"
as
a
test
to
p
ic
h
elp
ed
s
ee
h
o
w
well
it
wo
r
k
ed
.
I
n
s
tead
o
f
in
v
e
n
tin
g
id
ea
s
,
th
e
s
y
s
tem
p
u
lled
o
n
ly
f
r
o
m
w
h
at
th
e
r
ef
er
en
ce
o
f
f
er
e
d
.
E
ac
h
tim
e,
i
n
p
u
ts
wer
e
m
atch
e
d
ca
r
ef
u
lly
ag
ain
s
t e
x
is
tin
g
m
ater
ial.
L
o
o
k
at
Fig
u
r
e
4
-
b
asic
g
en
er
ativ
e
s
etu
p
s
,
wh
en
r
u
n
n
in
g
f
r
ee
o
f
R
AG
s
u
p
p
o
r
t,
to
s
s
ed
o
u
t
v
ag
u
e
ex
p
lan
atio
n
s
,
o
f
te
n
m
ix
in
g
u
p
C
++
’
s
ap
p
r
o
ac
h
to
p
o
l
y
m
o
r
p
h
is
m
with
h
o
w
Py
th
o
n
h
an
d
le
s
ty
p
es
o
n
th
e
f
ly
.
On
th
e
f
lip
s
id
e,
M
o
d
el
B
,
wh
ich
p
air
e
d
Mistra
l
7
B
with
R
AG,
p
u
lled
p
r
ec
is
e
ch
u
n
k
s
s
tr
aig
h
t
f
r
o
m
th
e
p
r
o
v
id
e
d
tex
tb
o
o
k
-
lik
e
p
ag
es
1
4
5
th
r
o
u
g
h
1
4
8
-
an
d
b
u
ilt
r
esp
o
n
s
es
u
s
in
g
th
e
v
er
y
wo
r
d
s
f
r
o
m
th
o
s
e
s
ec
tio
n
s
.
T
h
at
s
h
o
ws
th
e
C
h
r
o
m
aDB r
etr
iev
al
p
r
o
ce
s
s
ac
tu
al
ly
f
ix
es th
e
g
ap
m
en
tio
n
e
d
ea
r
lier
in
th
e
an
aly
s
is
.
Fig
u
r
e
4
.
C
o
m
p
ar
ativ
e
wo
r
k
f
l
o
w
an
aly
s
is
s
h
o
win
g
how
th
e
R
AG
p
ip
elin
e
b
r
id
g
es
th
e
co
n
t
ex
tu
al
d
is
co
n
n
ec
t
to
p
r
o
d
u
ce
alig
n
e
d
ed
u
ca
tio
n
a
l c
o
n
ten
t
5
.
2
.
Q
ua
ntit
a
t
iv
e
re
t
riev
a
l
e
f
f
i
c
a
c
y
A
s
co
r
e
ca
lled
h
it
r
ate
at
p
o
s
itio
n
th
r
ee
h
elp
e
d
ch
ec
k
h
o
w
w
ell
C
h
r
o
m
aDB
f
o
u
n
d
r
ele
v
an
t
in
f
o
.
On
e
f
in
d
s
it
b
y
s
ee
in
g
if
th
e
r
ig
h
t
tex
tb
o
o
k
p
ar
t
s
h
o
ws
u
p
am
o
n
g
th
e
f
ir
s
t
th
r
ee
r
esu
lts
p
u
lled
.
T
h
at
ch
an
ce
g
ets
wr
itten
d
o
wn
u
s
in
g
a
clea
r
m
a
th
r
u
le.
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
A
h
yb
r
id
r
etri
ev
a
l a
u
g
men
ted
g
en
era
tio
n
fr
a
mewo
r
k
fo
r
a
u
t
o
ma
ted
ed
u
ca
tio
n
a
l
…
(
B
a
s
a
v
esh
D.
)
1971
=
Star
tin
g
with
th
e
all
-
Min
iLM
-
L6
-
v
2
em
b
e
d
d
in
g
s
,
p
e
r
f
o
r
m
a
n
ce
r
ea
c
h
ed
a
n
8
9
%
h
it
r
ate
o
n
co
n
ce
p
t
-
b
ased
s
y
llab
u
s
s
ea
r
ch
es.
T
ak
e
th
e
p
h
r
ase
"E
x
p
lain
in
h
er
itan
ce
"
-
th
e
s
y
s
tem
p
u
lled
ju
s
t
th
e
s
ec
tio
n
d
etailin
g
o
b
ject
-
o
r
ien
te
d
in
h
er
itan
ce
.
I
n
s
tead
o
f
g
ettin
g
s
id
etr
ac
k
ed
b
y
co
m
m
o
n
wo
r
d
s
i
n
th
e
p
r
ef
ac
e
o
r
g
lo
s
s
ar
y
,
m
ea
n
in
g
-
d
r
iv
en
m
atch
i
n
g
s
tay
ed
f
o
cu
s
ed
.
B
ec
au
s
e
it
ig
n
o
r
ed
m
is
lead
in
g
k
e
y
wo
r
d
o
v
er
lap
s
,
r
elev
an
ce
im
p
r
o
v
e
d
s
h
ar
p
l
y
.
T
h
at
p
r
ec
is
io
n
s
h
o
ws
h
o
w
well
v
ec
to
r
s
e
ar
ch
p
in
p
o
in
ts
u
s
ef
u
l
co
n
tex
t
-
r
ea
d
y
i
n
g
ac
c
u
r
ate
in
p
u
ts
b
ef
o
r
e
th
e
L
L
M
s
tep
s
i
n
.
5
.
3
.
Q
ua
lit
a
t
iv
e
a
na
ly
s
is
:
Aut
o
m
a
t
ed
la
bo
ra
t
o
r
y
g
e
n
e
r
a
t
i
o
n
A
f
r
esh
lo
o
k
at
t
h
e
"L
ab
p
ip
e
lin
e
"
ch
ec
k
ed
h
o
w
well
it
h
a
n
d
les
m
ess
y
PDF
tex
t.
T
h
o
u
g
h
b
u
ilt
f
o
r
ch
ao
s
,
it
s
h
ap
es
th
at
clu
tter
in
to
clea
n
r
ep
o
r
ts
.
W
h
er
e
m
o
s
t
t
o
o
ls
f
ail,
th
is
o
n
e
p
u
s
h
es
th
r
o
u
g
h
-
tu
r
n
i
n
g
r
o
u
g
h
d
ata
in
to
s
o
m
eth
in
g
u
s
ab
le.
E
ac
h
s
tep
ad
ju
s
ts
o
n
th
e
f
ly
,
m
ak
in
g
s
en
s
e
o
f
wh
at
lo
o
k
s
lik
e
g
ib
b
er
is
h
.
No
t
p
er
f
ec
t,
y
et
cl
o
s
e
en
o
u
g
h
to
m
atter
.
E
v
e
n
o
d
d
f
o
r
m
attin
g
f
in
d
s
a
p
lace
i
n
th
e
o
u
tp
u
t.
W
h
en
test
ed
,
r
esu
lts
ca
m
e
o
u
t stru
ctu
r
e
d
,
n
o
t scr
a
m
b
led
.
C
lar
ity
em
er
g
es wh
er
e
y
o
u
’
d
ex
p
ec
t c
o
n
f
u
s
io
n
.
A
lab
g
u
id
e
s
er
v
ed
as th
e
test
d
o
cu
m
e
n
t.
I
n
s
id
e
it sat a
task
ab
o
u
t c
o
d
in
g
m
u
ltip
licatio
n
f
o
r
two
g
r
id
s
o
f
n
u
m
b
e
r
s
u
s
in
g
Py
th
o
n
.
T
h
i
s
in
s
tr
u
ctio
n
ap
p
ea
r
ed
alo
n
g
s
id
e
m
ess
y
ex
t
r
as
-
r
an
d
o
m
titl
es
f
r
o
m
ta
b
les,
m
ar
k
s
at
p
ag
e
b
o
tto
m
s
,
tim
e
tag
s
s
c
atter
ed
ar
o
u
n
d
.
W
h
at
s
to
o
d
o
u
t
was
h
o
w
clu
tter
ed
th
e
s
o
u
r
ce
m
ater
ial
f
elt
o
n
p
u
r
p
o
s
e.
No
t
clea
n
.
Far
f
r
o
m
n
ea
t.
Yet
th
at
s
p
ec
if
ic
q
u
esti
o
n
n
ee
d
ed
f
in
d
in
g
am
id
d
is
tr
ac
tio
n
s
.
On
e
p
ar
t
f
o
cu
s
ed
o
n
s
o
f
twar
e
s
tep
s
f
o
r
n
u
m
b
er
ar
r
ay
s
.
E
v
er
y
t
h
in
g
els
e
ju
s
t g
o
t in
th
e
way
.
Ou
t
o
f
th
e
m
ess
y
in
p
u
t,
a
p
att
er
n
-
b
ased
s
ca
n
n
er
p
u
lled
o
u
t
j
u
s
t
wh
at
was
n
ee
d
e
d
.
Fro
m
t
h
at
p
o
in
t
o
n
,
ev
er
y
th
in
g
ca
m
e
to
g
eth
er
in
clea
n
Ma
r
k
d
o
wn
lay
o
u
t.
T
h
is
in
clu
d
e
d
ea
c
h
r
e
q
u
ir
e
d
p
ar
t,
b
u
ilt
s
tep
b
y
s
tep
with
o
u
t e
x
tr
a
clu
tter
.
W
h
at
em
er
g
ed
f
it th
e
s
tr
u
ctu
r
e
ex
a
ctly
,
s
h
ap
ed
b
y
r
u
les b
u
t
n
o
t r
ig
i
d
A
s
in
g
le
r
o
w
at
a
tim
e
m
o
v
es
th
r
o
u
g
h
th
e
f
ir
s
t
m
atr
ix
.
M
ea
n
wh
ile,
ea
ch
co
lu
m
n
i
n
th
e
s
ec
o
n
d
is
p
r
o
ce
s
s
ed
b
esid
e
it.
On
e
ele
m
en
t
m
u
ltip
lies
a
n
o
th
er
b
ef
o
r
e
th
eir
r
esu
lt
ad
d
s
in
to
a
r
u
n
n
in
g
s
u
m
.
T
h
at
s
u
m
s
ettles
in
to
p
lace
o
n
ly
af
ter
all
p
air
s
co
n
tr
ib
u
te.
E
ac
h
p
o
s
itio
n
in
th
e
o
u
tco
m
e
f
ills
th
is
way
.
T
h
e
p
r
o
ce
s
s
r
ep
ea
ts
u
n
til ev
er
y
co
m
b
i
n
atio
n
co
m
p
letes
,
n
o
th
in
g
s
k
i
p
s
,
an
d
n
o
th
i
n
g
d
o
u
b
les.
Py
th
o
n
s
cr
ip
t
th
at
r
u
n
s
with
o
u
t
er
r
o
r
s
,
b
u
ilt
ex
ac
tly
to
m
atch
th
e
m
eth
o
d
ex
p
lain
e
d
.
E
x
ec
u
tio
n
-
r
ea
d
y
lin
es
o
f
co
d
e
f
o
llo
win
g
p
r
o
p
e
r
s
y
n
tax
r
u
les.
A
wo
r
k
i
n
g
p
r
o
g
r
am
wr
itten
in
Py
th
o
n
,
s
tr
u
c
tu
r
ed
co
r
r
ec
tly
f
o
r
im
m
ed
iate
u
s
e.
C
o
r
r
ec
tly
f
o
r
m
ed
in
s
tr
u
ctio
n
s
in
Py
th
o
n
ca
p
ab
le
o
f
r
u
n
n
in
g
as
in
ten
d
e
d
.
T
h
e
alg
o
r
ith
m
tu
r
n
ed
in
to
r
u
n
n
ab
le
co
d
e
with
ac
cu
r
ate
f
o
r
m
attin
g
.
T
im
e
g
r
o
ws
cu
b
e
d
,
s
p
ac
e
s
q
u
ar
ed
-
th
is
o
n
e
s
ca
le
s
teep
.
E
a
ch
s
tep
p
iles
u
p
,
m
em
o
r
y
f
ills
f
aster
th
an
y
o
u
’
d
f
ir
s
t
g
u
ess
.
T
r
ip
le
lo
o
p
s
d
r
ag
o
n
,
n
ested
d
ee
p
in
s
id
e
.
Me
m
o
r
y
tag
s
alo
n
g
,
g
r
o
win
g
twice
as
f
ast
as
in
p
u
t.
No
t le
an
,
n
o
t
q
u
ick
,
ju
s
t h
ea
v
y
liftin
g
all
r
o
u
n
d
.
Viv
a
q
u
esti
o
n
s
:
C
o
n
tex
tu
ally
r
elev
an
t
ass
ess
m
en
t
q
u
esti
o
n
s
(
e.
g
.
,
"Wh
at
co
n
d
itio
n
m
u
s
t
b
e
s
atis
f
ied
f
o
r
m
atr
ix
m
u
ltip
licatio
n
?"
a
n
d
"Can
y
o
u
o
p
tim
ize
m
atr
i
x
m
u
ltip
licatio
n
u
s
in
g
Stra
s
s
en
'
s
a
lg
o
r
ith
m
?"
)
.
A
s
in
g
le
test
s
h
o
ws
h
o
w
th
e
t
o
o
l
h
an
d
les
m
ess
y
f
iles
s
tar
t
to
f
in
is
h
.
Star
tin
g
with
u
n
clea
r
d
o
cu
m
en
ts
,
it
b
u
ild
s
clea
r
teac
h
in
g
m
ater
i
al.
I
n
s
tead
o
f
s
to
p
p
in
g
at
in
te
r
p
r
etatio
n
,
it
s
h
a
p
es
o
u
tp
u
t
th
at
f
its
less
o
n
g
o
als.
E
v
en
r
o
u
g
h
d
r
af
ts
tu
r
n
in
to
o
r
g
an
ized
s
ec
tio
n
s
r
ea
d
y
f
o
r
lab
d
o
cu
m
e
n
tatio
n
.
5
.
4
.
Q
ua
ntit
a
t
iv
e
c
o
m
pa
r
a
t
i
v
e
a
na
ly
s
is
(
a
bla
t
io
n
s
t
u
d
y
)
L
o
o
k
in
g
at
h
o
w
th
in
g
s
tu
r
n
e
d
o
u
t,
th
e
b
aselin
e
Mo
d
el
A
-
Fl
an
-
T
5
u
s
in
g
co
m
p
r
ess
ed
r
etr
ie
v
al
-
was
m
ea
s
u
r
ed
n
ex
t
t
o
Mo
d
el
B
,
w
h
ich
u
s
es
Mistra
l
7
B
p
lu
s
Fu
ll
-
C
o
n
tex
t
R
AG.
W
h
at
s
h
o
ws
u
p
in
T
ab
le
1
p
o
in
ts
to
clea
r
g
ap
s
b
etwe
en
th
em
o
n
ev
er
y
m
etr
ic
test
ed
.
Per
f
o
r
m
an
ce
is
n
o
t
clo
s
e
wh
en
y
o
u
g
o
th
r
o
u
g
h
ea
ch
m
ea
s
u
r
e
o
n
e
b
y
o
n
e.
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
m
e
t
r
i
c
s
F
e
a
t
u
r
e
M
o
d
e
l
A
(
B
a
s
e
l
i
n
e
:
F
l
a
n
-
T
5
)
M
o
d
e
l
B
(
P
r
o
p
o
s
e
d
:
M
i
st
r
a
l
7
B
)
A
r
c
h
i
t
e
c
t
u
r
e
E
n
c
o
d
e
r
-
D
e
c
o
d
e
r
(
8
0
M
P
a
r
a
m
s
)
D
e
c
o
d
e
r
-
O
n
l
y
(
7
B
P
a
r
a
m
s
)
C
o
n
t
e
x
t
S
t
r
a
t
e
g
y
Lo
ss
y
V
e
c
t
o
r
C
o
m
p
r
e
s
s
i
o
n
F
u
l
l
-
C
o
n
t
e
x
t
I
n
j
e
c
t
i
o
n
H
a
l
l
u
c
i
n
a
t
i
o
n
R
a
t
e
H
i
g
h
(
6
5
%)
-
I
n
v
e
n
t
e
d
n
o
n
-
e
x
i
st
e
n
t
A
P
I
s
.
Lo
w
(
1
2
%)
-
A
d
h
e
r
e
d
to
t
e
x
t
b
o
o
k
d
e
f
i
n
i
t
i
o
n
s
.
C
o
d
e
S
u
c
c
e
s
s
Rate
1
5
%
-
F
r
a
g
me
n
t
s
w
e
r
e
o
f
t
e
n
i
n
c
o
m
p
l
e
t
e
.
9
2
%
-
G
e
n
e
r
a
t
e
d
f
u
l
l
y
e
x
e
c
u
t
a
b
l
e
c
o
d
e
.
A
v
g
.
L
a
t
e
n
c
y
0
.
8
s
e
c
o
n
d
s
(
R
e
a
l
-
t
i
m
e
)
1
4
.
2
s
e
c
o
n
d
s
(
A
s
y
n
c
h
r
o
n
o
u
s
)
T
h
o
u
g
h
Mo
d
el
A
an
s
wer
s
f
ast
-
ju
s
t
0
.
8
s
ec
o
n
d
s
-
th
e
s
p
ee
d
co
m
es
at
a
s
teep
p
r
ice.
R
esp
o
n
s
es
o
f
ten
twis
t
m
ea
n
in
g
b
ey
o
n
d
r
ec
o
g
n
i
tio
n
,
th
an
k
s
to
its
co
m
p
r
ess
ed
way
o
f
p
u
llin
g
d
ata.
I
n
f
ac
t,
m
o
r
e
th
an
h
alf
th
e
o
u
tp
u
t
in
v
e
n
ts
d
etails
o
u
t
o
f
th
in
air
.
Fak
e
f
u
n
ctio
n
n
am
es
ap
p
ea
r
.
C
o
d
e
s
n
ip
p
ets
b
r
e
ak
b
asic
r
u
les.
Fo
r
s
ch
o
lar
ly
wo
r
k
,
wh
er
e
tr
u
th
m
atter
s
ab
o
v
e
all,
th
is
m
o
d
el
s
im
p
ly
ca
n
n
o
t b
e
tr
u
s
ted
.
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
1
96
4
-
1
975
1972
E
v
en
th
o
u
g
h
Mo
d
el
B
was
s
lo
wer
at
1
4
.
2
s
ec
o
n
d
s
,
it
h
it
t
h
e
m
in
im
u
m
u
s
ef
u
ln
ess
lev
el
n
ee
d
ed
to
wo
r
k
as
an
au
to
m
atic
teac
h
in
g
h
elp
er
.
B
ec
au
s
e
it
u
s
ed
a
d
ec
o
d
er
-
o
n
ly
s
etu
p
,
it
co
u
ld
ac
tu
a
lly
lear
n
o
n
th
e
f
ly
f
r
o
m
ex
am
p
les,
p
ick
in
g
u
p
co
d
in
g
p
atter
n
s
an
d
s
u
b
ject
ter
m
s
p
u
lled
s
tr
aig
h
t f
r
o
m
tex
tb
o
o
k
s
.
Sin
ce
9
2
p
er
ce
n
t
o
f
its
g
en
er
ated
co
d
e
r
an
co
r
r
ec
tly
,
th
e
m
eth
o
d
o
f
f
ee
d
in
g
f
u
ll
co
n
tex
t
k
ee
p
s
m
ea
n
in
g
in
t
ac
t
well
en
o
u
g
h
f
o
r
co
n
s
is
ten
t g
r
ad
in
g
b
en
ch
m
ar
k
s
lik
e
m
ass
iv
e
m
u
ltit
ask
lan
g
u
ag
e
u
n
d
er
s
tan
d
in
g
[
2
5
]
.
5
.
5
.
Dis
cus
s
io
n
on
pr
iv
a
cy
a
nd
l
a
t
e
n
c
y
On
e
b
ig
h
u
r
d
le
h
e
r
e
is
k
ee
p
in
g
ev
er
y
b
it
o
f
d
ata
f
u
lly
u
n
d
e
r
co
n
tr
o
l
at
ea
ch
s
tep
.
T
h
e
wh
o
le
p
r
o
ce
s
s
h
ad
to
r
esp
ec
t th
at
lim
it with
o
u
t e
x
ce
p
tio
n
.
Fo
u
r
teen
s
ec
o
n
d
s
–
th
at
i
s
h
o
w
lo
n
g
it
tak
es
a
r
eg
u
lar
lap
t
o
p
with
1
6
GB
R
AM
b
u
t
n
o
ex
tr
a
g
r
ap
h
ics
ca
r
d
to
p
r
o
ce
s
s
o
n
e
r
eq
u
est.
C
lo
u
d
s
y
s
tem
s
o
f
ten
f
in
is
h
in
u
n
d
er
two
s
ec
o
n
d
s
,
m
u
ch
f
as
ter
b
y
co
m
p
a
r
is
o
n
.
Sti
ll,
wh
en
lear
n
er
s
g
o
th
r
o
u
g
h
less
o
n
s
at
th
eir
o
wn
p
ac
e,
waitin
g
is
n
o
t
an
is
s
u
e.
Stu
d
y
p
lan
s
an
d
co
u
r
s
e
o
u
tlin
es c
an
wait
-
th
e
f
l
o
w
s
tay
s
s
tead
y
with
o
u
t n
ee
d
in
g
in
s
t
an
t r
ep
lies
.
Kee
p
in
g
th
in
g
s
p
r
iv
ate
m
atter
s
m
o
r
e
th
an
m
o
s
t
ad
m
it.
I
n
s
te
ad
o
f
s
en
d
i
n
g
co
u
r
s
e
d
etails
o
f
f
to
d
is
tan
t
co
m
p
u
ter
s
lik
e
th
o
s
e
b
ig
tech
s
y
s
tem
s
o
f
ten
d
o
,
ev
er
y
th
in
g
r
u
n
s
in
s
id
e
y
o
u
r
o
wn
n
etwo
r
k
h
er
e.
T
h
at
m
ea
n
s
an
y
th
in
g
d
elica
te
-
s
ay
,
u
n
f
in
i
s
h
ed
less
o
n
p
la
n
s
o
r
in
ter
n
al
ex
p
er
im
en
t
g
u
i
d
es
-
n
ev
er
lea
v
es
ca
m
p
u
s
walls.
W
h
at
s
its
b
eh
in
d
th
is
s
etu
p
is
n
o
t
m
ag
ic,
ju
s
t
ca
r
e
f
u
l
d
es
ig
n
.
Sch
o
o
ls
b
o
u
n
d
b
y
tig
h
t
r
u
les
o
n
h
a
n
d
lin
g
in
f
o
r
m
atio
n
ten
d
t
o
f
in
d
th
is
way
less
s
tr
es
s
f
u
l.
5
.
6
.
Cha
lleng
es
a
nd
l
i
mi
t
a
t
i
o
n
s
Sti
ll,
a
h
an
d
f
u
l
o
f
g
litch
es
p
o
p
p
ed
u
p
d
u
r
i
n
g
tr
ials
,
d
esp
ite
p
r
o
o
f
th
at
l
o
ca
l
R
AG
s
y
s
tem
s
f
u
n
ctio
n
in
class
r
o
o
m
s
.
T
h
in
g
s
d
id
n
o
t
u
n
f
o
ld
with
o
u
t
h
iccu
p
s
o
n
ce
u
s
e
d
o
u
ts
id
e
co
n
tr
o
lled
s
ettin
g
s
.
5
.
6
.
1
.
Do
cum
ent
s
t
ruct
ure
v
a
r
i
e
s
A
s
in
g
le
m
ajo
r
s
n
ag
with
in
th
e
I
n
g
esti
o
n
L
ay
er
lies
in
h
o
w
ea
ch
u
n
iv
e
r
s
ity
ar
r
an
g
es
its
PD
F
s
tr
u
ctu
r
e
th
eir
o
wn
way
.
W
h
ile
Py
Mu
PDF
ex
tr
ac
ts
r
aw
in
f
o
r
m
atio
n
,
s
p
ec
ially
d
esig
n
ed
r
u
les
attem
p
t
in
ter
p
r
etatio
n
-
b
u
t
to
g
eth
er
th
ey
f
r
e
q
u
en
tly
f
ail.
E
s
p
ec
ially
wh
en
d
o
cu
m
en
ts
s
h
if
t
in
to
m
u
lti
-
co
lu
m
n
lay
o
u
ts
.
Or
b
u
r
y
class
d
etails
d
ee
p
in
s
id
e
co
m
p
le
x
tab
le
s
h
ap
es
r
ath
er
th
an
clea
r
r
o
ws.
T
h
e
n
co
n
f
u
s
io
n
s
p
r
ea
d
s
f
ast.
Fin
d
in
g
n
o
th
in
g
tag
g
ed
Mo
d
u
le
-
3
?
Pieces
v
an
is
h
o
u
t
o
f
n
o
wh
er
e.
T
h
o
s
e
em
p
ty
s
p
o
ts
s
h
o
w
u
p
lo
u
d
an
d
clea
r
wh
en
th
in
g
s
f
all
a
p
ar
t m
u
c
h
la
ter
.
5
.
6
.
2
.
L
o
s
s
of
m
ea
nin
g
a
cr
o
s
s
m
ultiple
mo
d
e
s
Mo
s
t
lear
n
in
g
p
a
g
es
p
air
tex
t
alo
n
g
s
id
e
s
k
etch
es
-
cir
c
u
its
,
f
lo
w
m
ap
s
,
n
u
m
b
er
s
ig
n
s
-
n
o
t
ju
s
t
p
ar
ag
r
a
p
h
s
alo
n
e.
At
p
r
esen
t,
th
e
s
o
f
twar
e
co
n
v
er
tin
g
d
o
c
u
m
en
ts
in
to
s
ea
r
ch
-
r
ea
d
y
m
at
er
ial
p
ick
s
u
p
o
n
ly
wr
itten
lin
es,
ig
n
o
r
in
g
ea
ch
v
is
u
al
co
m
p
letely
d
u
r
in
g
s
to
r
ag
e.
As
a
r
esu
lt,
q
u
esti
o
n
s
ti
ed
to
im
ag
es
-
f
o
r
ex
am
p
le,
ex
p
lain
in
g
th
e
lay
o
u
t
s
h
o
wn
o
n
p
ag
e
4
5
-
g
et
n
o
r
esp
o
n
s
e
wh
atso
ev
er
.
T
h
is
g
ap
,
th
o
u
g
h
s
ilen
t,
s
tr
etch
es f
ar
: w
h
at
h
u
m
an
s
u
n
d
er
s
tan
d
s
tay
s
r
ich
,
wh
ile
m
ac
h
in
e
r
ec
o
r
d
s
s
tay
th
in
.
5
.
6
.
3
.
Co
m
pu
t
a
t
i
o
na
l
lim
it
s
a
nd
d
el
a
y
s
Stay
in
g
in
ch
a
r
g
e
o
f
in
f
o
r
m
at
io
n
m
ea
n
t
r
u
n
n
in
g
e
v
er
y
th
i
n
g
lo
ca
lly
,
u
s
in
g
Ollam
a
alo
n
g
s
i
d
e
Mistra
l
7
B
.
E
v
en
s
o
,
p
e
r
f
o
r
m
an
ce
s
lo
ws
d
o
wn
q
u
ite
a
b
it.
Usi
n
g
c
o
m
m
o
n
h
ar
d
war
e
-
a
s
tan
d
a
r
d
i7
ch
i
p
,
1
6
GB
o
f
R
AM
-
f
u
ll
r
ep
lies
ty
p
ically
n
ee
d
ab
o
u
t
f
o
u
r
teen
s
ec
o
n
d
s
.
T
h
at
tim
in
g
wo
r
k
s
f
in
e
wh
en
ex
p
lo
r
in
g
id
ea
s
alo
n
e,
o
u
ts
id
e
an
y
r
u
s
h
.
B
u
t
s
u
c
h
d
e
lay
s
m
ak
e
b
ac
k
-
an
d
-
f
o
r
th
co
n
v
er
s
atio
n
s
f
ee
l
awk
war
d
.
On
li
n
e
s
er
v
ices
an
s
wer
q
u
ick
er
,
f
r
eq
u
en
tly
in
less
th
a
n
two
s
ec
o
n
d
s
.
C
lear
ly
,
d
is
co
n
n
ec
ted
to
o
ls
ca
n
n
o
t
m
atch
t
h
at
p
ac
e.
5
.
6
.
4
.
Co
nte
x
t
wind
o
w
b
o
t
t
l
en
e
c
ks
Desp
ite
h
an
d
lin
g
8
k
t
o
k
en
s
,
Mistra
l
7
B
q
u
ick
ly
u
s
es
u
p
s
p
ac
e
wh
en
g
r
a
b
b
in
g
ch
u
n
k
s
o
f
tex
tb
o
o
k
s
v
ia
R
AG.
Har
d
war
e
co
n
s
tr
ain
ts
allo
w
ju
s
t
th
r
ee
to
p
-
m
at
ch
in
g
s
ec
tio
n
s
p
er
q
u
er
y
.
I
f
a
co
n
ce
p
t
s
u
c
h
as
Dy
n
am
ic
Pro
g
r
am
m
i
n
g
s
p
an
s
m
u
ltip
le
ch
ap
ter
s
,
o
r
p
u
lls
in
f
o
f
r
o
m
f
a
r
en
d
s
o
f
th
e
m
ater
i
al,
th
ese
f
r
ag
m
en
ts
m
ay
s
k
ip
ess
en
tial
s
etu
p
.
An
s
wer
s
th
en
s
o
u
n
d
th
in
-
m
is
s
in
g
co
n
n
ec
tio
n
s
th
ey
o
u
g
h
t
t
o
in
clu
d
e.
W
ith
o
u
t
en
o
u
g
h
s
u
r
r
o
u
n
d
in
g
d
etail,
d
e
p
th
f
ad
es.
C
o
n
tex
t
r
u
n
n
in
g
l
o
w
d
o
es th
at.
6.
CO
NCLU
SI
O
N
Her
e
co
m
es
th
e
s
etu
p
-
cr
af
ted
a
s
ec
u
r
e
s
y
s
tem
to
alig
n
co
u
r
s
e
p
lan
s
with
tex
tb
o
o
k
m
at
er
ial,
u
s
in
g
s
m
ar
t
r
etr
iev
al
m
eth
o
d
s
tied
t
o
co
n
ten
t
cr
ea
tio
n
.
A
b
r
i
d
g
e
f
o
r
m
ed
w
h
er
e
s
tr
ict
ac
ad
em
ic
o
u
tlin
es
m
et
b
r
o
ad
r
ea
d
in
g
s
o
u
r
c
es,
h
an
d
led
p
r
iv
ately
.
T
h
e
m
eth
o
d
lea
n
ed
o
n
en
h
an
ce
d
g
en
er
atio
n
m
o
d
els,
p
u
llin
g
f
ac
ts
wh
en
n
ee
d
ed
in
s
tead
o
f
r
ely
in
g
s
o
lely
o
n
p
r
eset
r
esp
o
n
s
es.
Secu
r
ity
s
tay
ed
in
tact
th
r
o
u
g
h
o
u
t
ea
c
h
s
tep
,
ev
en
d
u
r
in
g
h
ea
v
y
co
m
p
ar
is
o
n
s
.
Ma
tch
in
g
ac
cu
r
ac
y
im
p
r
o
v
e
d
b
ec
au
s
e
c
o
n
tex
t
m
atter
ed
m
o
r
e
th
an
e
v
er
b
ef
o
r
e
.
Stru
ctu
r
e
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
A
h
yb
r
id
r
etri
ev
a
l a
u
g
men
ted
g
en
era
tio
n
fr
a
mewo
r
k
fo
r
a
u
t
o
ma
ted
ed
u
ca
tio
n
a
l
…
(
B
a
s
a
v
esh
D.
)
1973
f
o
llo
wed
p
u
r
p
o
s
e,
y
et
f
lex
ib
il
ity
r
em
ain
ed
with
in
r
ea
ch
.
T
esti
n
g
s
h
o
wed
co
n
s
is
ten
t
r
esu
lts
ac
r
o
s
s
d
if
f
er
en
t
s
u
b
jects.
Priv
ac
y
d
id
n
o
t
s
lip
,
d
esp
ite
co
m
p
lex
p
r
o
ce
s
s
in
g
d
em
an
d
s
.
Per
f
o
r
m
a
n
ce
h
eld
s
tead
y
u
n
d
er
r
ea
l
co
n
d
itio
n
s
.
No
t e
v
er
y
a
p
p
r
o
ac
h
wo
r
k
s
th
is
way
-
b
u
t th
is
o
n
e
d
id
.
I
t
tu
r
n
s
o
u
t
s
p
ee
d
is
n
o
t
e
v
e
r
y
th
in
g
wh
e
n
teac
h
in
g
m
ac
h
i
n
es
to
ass
is
t
lear
n
in
g
.
T
h
o
u
g
h
Flan
-
T
5
r
esp
o
n
d
s
q
u
ick
ly
,
it
o
f
ten
in
v
e
n
ts
d
etails
-
ab
o
u
t
two
th
i
r
d
s
o
f
th
e
tim
e
-
wh
ic
h
m
ak
es
it
u
n
r
eliab
le
f
o
r
s
ch
o
o
l
wo
r
k
.
I
n
s
tead
o
f
b
alan
cin
g
s
p
ee
d
an
d
ac
cu
r
ac
y
,
we
f
o
u
n
d
a
d
if
f
e
r
en
t
p
at
h
wo
r
k
s
b
etter
.
B
y
f
ee
d
in
g
f
u
l
l
co
n
tex
t
th
r
o
u
g
h
C
h
r
o
m
aDB
in
to
Mistra
l
7
B
,
co
r
r
ec
t
o
u
tp
u
ts
r
o
s
e
s
h
ar
p
ly
.
Mo
s
t
attem
p
ts
m
atch
ed
cu
r
r
icu
l
u
m
r
u
les,
h
itti
n
g
a
9
2
p
er
ce
n
t
m
ar
k
with
o
u
t
clo
u
d
s
u
p
p
o
r
t.
H
ar
d
war
e
lim
its
d
id
n
o
t
s
to
p
s
o
lid
r
esu
lts
o
n
r
ea
l
class
r
o
o
m
-
s
ty
le
p
r
o
b
lem
s
.
W
h
at
lo
o
k
e
d
lik
e
a
tr
ad
e
-
o
f
f
at
f
ir
s
t
tu
r
n
ed
in
to
a
clea
r
lead
f
o
r
o
n
e
d
esig
n
.
Hea
v
y
r
elian
ce
o
n
m
em
o
r
y
-
r
ich
g
e
n
er
atio
n
b
ea
t
s
h
o
r
tcu
ts
ev
er
y
tim
e.
No
t
all
s
m
all
s
y
s
tem
s
th
i
n
k
alik
e
-
o
n
e
k
in
d
s
im
p
ly
u
n
d
e
r
s
tan
d
s
b
etter
.
F
r
o
m
h
e
r
e
o
n
,
s
c
h
o
o
l
s
c
a
n
b
u
i
l
d
t
h
e
i
r
o
w
n
s
m
a
r
t
h
e
l
p
e
r
s
u
s
i
n
g
c
l
e
a
r
s
t
e
p
s
a
n
y
o
n
e
c
o
u
l
d
f
o
l
l
o
w
.
T
h
e
s
e
t
o
o
l
s
r
u
n
b
y
t
h
e
m
s
el
v
e
s
y
et
s
ta
y
r
e
l
i
a
b
l
e
w
i
t
h
o
u
t
d
r
a
i
n
i
n
g
b
u
d
g
e
t
s
.
A
c
c
u
r
a
c
y
h
o
l
d
s
s
t
r
o
n
g
w
h
i
l
e
p
r
i
v
a
c
y
s
t
a
y
s
f
u
l
l
y
u
n
d
e
r
s
c
h
o
o
l
c
o
n
t
r
o
l
.
W
h
a
t
m
a
t
t
e
r
s
m
o
s
t
is
t
o
t
al
o
w
n
e
r
s
h
i
p
o
f
i
n
f
o
r
m
a
t
i
o
n
-
n
o
e
x
c
e
p
t
i
o
n
s
e
v
e
r
.
7.
F
UT
UR
E
SCO
P
E
Fin
d
in
g
o
u
t
h
o
w
well
lo
ca
l
R
AG
wo
r
k
s
n
o
w
o
p
en
s
u
p
n
ew
p
ath
s
a
h
ea
d
.
So
m
e
g
ap
s
s
till
n
ee
d
cl
o
s
er
lo
o
k
s
later
o
n
.
W
ay
s
f
o
r
war
d
in
clu
d
e
test
in
g
b
r
o
ad
e
r
d
atas
ets
in
s
tead
o
f
n
ar
r
o
w
o
n
es.
E
x
p
lo
r
in
g
d
i
f
f
er
en
t
m
o
d
els
m
ig
h
t
r
ev
ea
l
h
i
d
d
en
p
atter
n
s
to
o
.
Ad
ju
s
tin
g
r
etr
iev
a
l
m
eth
o
d
s
co
u
ld
c
h
an
g
e
o
u
tco
m
es
s
h
ar
p
ly
.
E
v
en
h
ar
d
war
e
ch
o
ices
m
ay
in
f
lu
e
n
ce
r
esu
lts
m
o
r
e
th
an
e
x
p
ec
t
ed
.
E
ac
h
s
tep
u
n
co
v
er
s
q
u
esti
o
n
s
th
at
wer
e
n
o
t
o
b
v
io
u
s
at
f
i
r
s
t
Pictu
r
e
-
b
ased
r
ea
s
o
n
in
g
co
u
ld
s
tep
in
wh
er
e
d
ata
s
lip
s
th
r
o
u
g
h
g
a
p
s
,
as
s
ee
n
ea
r
lier
.
I
n
s
t
ea
d
of
ju
s
t
wo
r
d
s
,
to
o
ls
lik
e
L
L
aVA
m
i
g
h
t
h
el
p
d
ec
o
d
e
v
is
u
als
-
s
ay
,
wir
in
g
s
k
etch
es
o
r
m
ath
f
o
r
m
u
las
-
in
co
m
in
g
u
p
d
ates.
Seein
g
d
iag
r
am
s
m
ak
es
id
ea
s
clea
r
er
th
a
n
tex
t
alo
n
e.
L
ea
r
n
i
n
g
g
ets
r
ich
er
wh
en
b
o
th
e
y
es
an
d
m
in
d
en
g
ag
e
t
o
g
e
t
h
e
r
.
A
f
r
esh
s
ea
r
ch
m
et
h
o
d
tak
e
s
s
h
ap
e
wh
en
m
ea
n
in
g
-
b
ased
r
esu
lts
m
ee
t
ter
m
-
f
o
cu
s
ed
m
atch
in
g
.
I
n
s
tead
o
f
r
el
y
in
g
o
n
o
n
e
a
p
p
r
o
ac
h
alo
n
e
,
th
e
s
etu
p
u
s
es
b
o
th
v
ec
t
o
r
p
atter
n
s
alo
n
g
s
i
d
e
tr
ad
itio
n
al
wo
r
d
s
ig
n
als
lik
e
B
M2
5
.
W
h
en
odd
ac
r
o
n
y
m
s
s
h
o
w
up,
or
p
r
ec
is
e
co
d
e
lab
els
ap
p
ea
r
,
r
esp
o
n
s
es
s
tan
d
a
b
etter
ch
an
ce
of
h
itti
n
g
clo
s
e.
Sp
ec
i
f
ic
ter
m
s
g
ain
clar
ity
b
ec
a
u
s
e
r
aw
k
ey
wo
r
d
s
s
tep
in
w
h
er
e
c
o
n
tex
t
m
ig
h
t
m
is
s
.
Acc
u
r
ac
y
s
h
if
ts
u
p
war
d
o
n
ce
th
ese
two
s
ty
les r
u
n
to
g
eth
e
r
b
eh
in
d
th
e
s
ce
n
es.
On
e
way
f
o
r
war
d
?
B
u
ild
a
k
n
o
wled
g
e
g
r
ap
h
th
at
f
o
llo
ws
ea
ch
s
tu
d
en
t'
s
s
tep
s
.
T
h
is
p
ath
lets
th
e
AI
s
p
o
t
s
h
ak
y
s
p
o
ts
in
u
n
d
e
r
s
tan
d
in
g
d
u
r
in
g
s
p
o
k
e
n
-
s
ty
le
q
u
iz
ze
s
.
As
it
watc
h
es,
th
e
s
y
s
te
m
s
h
if
ts
-
n
o
l
o
n
g
er
ju
s
t
r
ep
ly
in
g
,
b
u
t
g
u
id
in
g
.
I
n
s
tead
o
f
waitin
g
,
it
s
u
g
g
ests
wh
at
to
tack
le
n
e
x
t.
Pro
g
r
ess
m
ea
n
s
m
o
v
in
g
f
r
o
m
s
tatic
an
s
wer
s
to
war
d
tailo
r
ed
s
u
p
p
o
r
t.
E
ac
h
in
ter
ac
tio
n
s
h
a
p
es
th
e
n
ex
t
n
u
d
g
e
.
Gr
o
wth
h
ap
p
en
s
b
y
lear
n
in
g
wh
o
s
tr
u
g
g
les wh
er
e.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
th
an
k
J
y
o
th
y
I
n
s
titu
te
of
T
ec
h
n
o
lo
g
y
f
o
r
p
r
o
v
id
i
n
g
s
u
p
p
o
r
t
an
d
i
n
f
r
a
s
t
r
u
c
t
u
r
e
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
d
id
not
r
ec
eiv
e
a
n
y
s
p
ec
if
ic
g
r
a
n
t
or
c
o
n
tr
ac
t
f
r
o
m
f
u
n
d
in
g
a
g
e
n
c
i
e
s
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
B
as
a
v
e
s
h
D.
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
J
a
y
a
s
h
r
e
e
N
a
g
a
r
aj
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.