<s>
Synthetic	B-General_Concept
data	I-General_Concept
is	O
information	O
that	O
's	O
artificially	O
generated	O
rather	O
than	O
produced	O
by	O
real-world	O
events	O
.	O
</s>
<s>
Typically	O
created	O
using	O
algorithms	O
,	O
synthetic	B-General_Concept
data	I-General_Concept
can	O
be	O
deployed	O
to	O
validate	O
mathematical	O
models	O
and	O
to	O
train	O
machine	O
learning	O
models	O
.	O
</s>
<s>
Data	O
generated	O
by	O
a	O
computer	O
simulation	O
can	O
be	O
seen	O
as	O
synthetic	B-General_Concept
data	I-General_Concept
.	O
</s>
<s>
Synthetic	B-General_Concept
data	I-General_Concept
is	O
used	O
in	O
a	O
variety	O
of	O
fields	O
as	O
a	O
filter	O
for	O
information	O
that	O
would	O
otherwise	O
compromise	O
the	O
confidentiality	O
of	O
particular	O
aspects	O
of	O
the	O
data	O
.	O
</s>
<s>
In	O
many	O
sensitive	O
applications	O
,	O
datasets	B-General_Concept
theoretically	O
exist	O
but	O
cannot	O
be	O
released	O
to	O
the	O
general	O
public	O
;	O
synthetic	B-General_Concept
data	I-General_Concept
sidesteps	O
the	O
privacy	O
issues	O
that	O
arise	O
from	O
using	O
real	O
consumer	O
information	O
without	O
permission	O
or	O
compensation	O
.	O
</s>
<s>
Synthetic	B-General_Concept
data	I-General_Concept
is	O
generated	O
to	O
meet	O
specific	O
needs	O
or	O
certain	O
conditions	O
that	O
may	O
not	O
be	O
found	O
in	O
the	O
original	O
,	O
real	O
data	O
.	O
</s>
<s>
This	O
can	O
be	O
useful	O
when	O
designing	O
any	O
type	O
of	O
system	O
because	O
the	O
synthetic	B-General_Concept
data	I-General_Concept
are	O
used	O
as	O
a	O
simulation	O
or	O
as	O
a	O
theoretical	O
value	O
,	O
situation	O
,	O
etc	O
.	O
</s>
<s>
Synthetic	B-General_Concept
data	I-General_Concept
are	O
often	O
generated	O
to	O
represent	O
the	O
authentic	O
data	O
and	O
allows	O
a	O
baseline	O
to	O
be	O
set	O
.	O
</s>
<s>
Another	O
benefit	O
of	O
synthetic	B-General_Concept
data	I-General_Concept
is	O
to	O
protect	O
the	O
privacy	O
and	O
confidentiality	O
of	O
authentic	O
data	O
.	O
</s>
<s>
As	O
stated	O
previously	O
,	O
synthetic	B-General_Concept
data	I-General_Concept
is	O
used	O
in	O
testing	O
and	O
creating	O
many	O
different	O
types	O
of	O
systems	O
;	O
below	O
is	O
a	O
quote	O
from	O
the	O
abstract	O
of	O
an	O
article	O
that	O
describes	O
a	O
software	O
that	O
generates	O
synthetic	B-General_Concept
data	I-General_Concept
for	O
testing	O
fraud	O
detection	O
systems	O
that	O
further	O
explains	O
its	O
use	O
and	O
importance	O
.	O
</s>
<s>
For	O
example	O
,	O
research	O
into	O
synthesis	O
of	O
audio	O
and	O
voice	B-Application
can	O
be	O
traced	O
back	O
to	O
the	O
1930s	O
and	O
before	O
,	O
driven	O
forward	O
by	O
the	O
developments	O
of	O
e.g.	O
</s>
<s>
In	O
the	O
context	O
of	O
privacy-preserving	O
statistical	O
analysis	O
,	O
in	O
1993	O
,	O
the	O
idea	O
of	O
original	O
fully	O
synthetic	B-General_Concept
data	I-General_Concept
was	O
created	O
by	O
Rubin	O
.	O
</s>
<s>
Later	O
that	O
year	O
,	O
the	O
idea	O
of	O
original	O
partially	O
synthetic	B-General_Concept
data	I-General_Concept
was	O
created	O
by	O
Little	O
.	O
</s>
<s>
Collectively	O
they	O
came	O
up	O
with	O
a	O
solution	O
for	O
how	O
to	O
treat	O
partially	O
synthetic	B-General_Concept
data	I-General_Concept
with	O
missing	O
data	O
.	O
</s>
<s>
Similarly	O
they	O
came	O
up	O
with	O
the	O
technique	O
of	O
Sequential	O
Regression	O
Multivariate	O
Imputation	B-General_Concept
.	O
</s>
<s>
Researchers	O
test	O
the	O
framework	O
on	O
synthetic	B-General_Concept
data	I-General_Concept
,	O
which	O
is	O
"	O
the	O
only	O
source	O
of	O
ground	O
truth	O
on	O
which	O
they	O
can	O
objectively	O
assess	O
the	O
performance	O
of	O
their	O
algorithms	O
"	O
.	O
</s>
<s>
Synthetic	B-General_Concept
data	I-General_Concept
can	O
be	O
generated	O
through	O
the	O
use	O
of	O
random	O
lines	O
,	O
having	O
different	O
orientations	O
and	O
starting	O
positions	O
.	O
</s>
<s>
Datasets	B-General_Concept
can	O
get	O
fairly	O
complicated	O
.	O
</s>
<s>
A	O
more	O
complicated	O
dataset	B-General_Concept
can	O
be	O
generated	O
by	O
using	O
a	O
synthesizer	O
build	O
.	O
</s>
<s>
In	O
a	O
linear	B-General_Concept
regression	I-General_Concept
line	B-General_Concept
example	O
,	O
the	O
original	O
data	O
can	O
be	O
plotted	O
,	O
and	O
a	O
best	O
fit	O
linear	B-General_Concept
line	I-General_Concept
can	O
be	O
created	O
from	O
the	O
data	O
.	O
</s>
<s>
This	O
line	B-General_Concept
is	O
a	O
synthesizer	O
created	O
from	O
the	O
original	O
data	O
.	O
</s>
<s>
The	O
next	O
step	O
will	O
be	O
generating	O
more	O
synthetic	B-General_Concept
data	I-General_Concept
from	O
the	O
synthesizer	O
build	O
or	O
from	O
this	O
linear	B-General_Concept
line	I-General_Concept
equation	O
.	O
</s>
<s>
David	O
Jensen	O
from	O
the	O
Knowledge	O
Discovery	O
Laboratory	O
explains	O
how	O
to	O
generate	O
synthetic	B-General_Concept
data	I-General_Concept
:	O
"	O
Researchers	O
frequently	O
need	O
to	O
explore	O
the	O
effects	O
of	O
certain	O
data	O
characteristics	O
on	O
their	O
data	O
model.	O
"	O
</s>
<s>
To	O
help	O
construct	O
datasets	B-General_Concept
exhibiting	O
specific	O
properties	O
,	O
such	O
as	O
auto-correlation	O
or	O
degree	O
disparity	O
,	O
proximity	O
can	O
generate	O
synthetic	B-General_Concept
data	I-General_Concept
having	O
one	O
of	O
several	O
types	O
of	O
graph	B-Application
structure	I-Application
:	O
random	O
graphs	O
that	O
are	O
generated	O
by	O
some	O
random	O
process	O
;	O
lattice	O
graphs	O
having	O
a	O
ring	O
structure	O
;	O
lattice	O
graphs	O
having	O
a	O
grid	O
structure	O
,	O
etc	O
.	O
</s>
<s>
Generate	O
the	O
empty	O
graph	B-Application
structure	I-Application
.	O
</s>
<s>
Testing	O
and	O
training	O
fraud	O
detection	O
and	O
confidentiality	O
systems	O
are	O
devised	O
using	O
synthetic	B-General_Concept
data	I-General_Concept
.	O
</s>
<s>
For	O
example	O
,	O
intrusion	O
detection	O
software	O
is	O
tested	O
using	O
synthetic	B-General_Concept
data	I-General_Concept
.	O
</s>
<s>
The	O
synthetic	B-General_Concept
data	I-General_Concept
allows	O
the	O
software	O
to	O
recognize	O
these	O
situations	O
and	O
react	O
accordingly	O
.	O
</s>
<s>
If	O
synthetic	B-General_Concept
data	I-General_Concept
was	O
not	O
used	O
,	O
the	O
software	O
would	O
only	O
be	O
trained	O
to	O
react	O
to	O
the	O
situations	O
provided	O
by	O
the	O
authentic	O
data	O
and	O
it	O
may	O
not	O
recognize	O
another	O
type	O
of	O
intrusion	O
.	O
</s>
<s>
Researchers	O
doing	O
clinical	O
trials	O
or	O
any	O
other	O
research	O
may	O
generate	O
synthetic	B-General_Concept
data	I-General_Concept
to	O
aid	O
in	O
creating	O
a	O
baseline	O
for	O
future	O
studies	O
and	O
testing	O
.	O
</s>
<s>
Real	O
data	O
can	O
contain	O
information	O
that	O
researchers	O
may	O
not	O
want	O
released	O
,	O
so	O
synthetic	B-General_Concept
data	I-General_Concept
is	O
sometimes	O
used	O
to	O
protect	O
the	O
privacy	O
and	O
confidentiality	O
of	O
a	O
dataset	B-General_Concept
.	O
</s>
<s>
Using	O
synthetic	B-General_Concept
data	I-General_Concept
reduces	O
confidentiality	O
and	O
privacy	O
issues	O
since	O
it	O
holds	O
no	O
personal	O
information	O
and	O
cannot	O
be	O
traced	O
back	O
to	O
any	O
individual	O
.	O
</s>
<s>
Synthetic	B-General_Concept
data	I-General_Concept
is	O
increasingly	O
being	O
used	O
for	O
machine	O
learning	O
applications	O
:	O
a	O
model	O
is	O
trained	O
on	O
a	O
synthetically	O
generated	O
dataset	B-General_Concept
with	O
the	O
intention	O
of	O
transfer	B-General_Concept
learning	I-General_Concept
to	O
real	O
data	O
.	O
</s>
<s>
Efforts	O
have	O
been	O
made	O
to	O
construct	O
general-purpose	O
synthetic	B-General_Concept
data	I-General_Concept
generators	O
to	O
enable	O
data	O
science	O
experiments	O
.	O
</s>
<s>
In	O
general	O
,	O
synthetic	B-General_Concept
data	I-General_Concept
has	O
several	O
natural	O
advantages	O
:	O
</s>
<s>
synthetic	B-General_Concept
data	I-General_Concept
can	O
have	O
perfectly	O
accurate	O
labels	O
,	O
including	O
labeling	O
that	O
may	O
be	O
very	O
expensive	O
or	O
impossible	O
to	O
obtain	O
by	O
hand	O
;	O
</s>
<s>
synthetic	B-General_Concept
data	I-General_Concept
can	O
be	O
used	O
as	O
a	O
substitute	O
for	O
certain	O
real	O
data	O
segments	O
that	O
contain	O
,	O
e.g.	O
,	O
sensitive	O
information	O
.	O
</s>
<s>
This	O
usage	O
of	O
synthetic	B-General_Concept
data	I-General_Concept
has	O
been	O
proposed	O
for	O
computer	O
vision	O
applications	O
,	O
in	O
particular	O
object	B-General_Concept
detection	I-General_Concept
,	O
where	O
the	O
synthetic	O
environment	O
is	O
a	O
3D	O
model	O
of	O
the	O
object	O
,	O
and	O
learning	O
to	O
navigate	O
environments	O
by	O
visual	O
information	O
.	O
</s>
<s>
At	O
the	O
same	O
time	O
,	O
transfer	B-General_Concept
learning	I-General_Concept
remains	O
a	O
nontrivial	O
problem	O
,	O
and	O
synthetic	B-General_Concept
data	I-General_Concept
has	O
not	O
become	O
ubiquitous	O
yet	O
.	O
</s>
<s>
Research	O
results	O
indicate	O
that	O
adding	O
a	O
small	O
amount	O
of	O
real	O
data	O
significantly	O
improves	O
transfer	B-General_Concept
learning	I-General_Concept
with	O
synthetic	B-General_Concept
data	I-General_Concept
.	O
</s>
<s>
Advances	O
in	O
generative	B-Algorithm
adversarial	I-Algorithm
networks	I-Algorithm
(	O
GAN	O
)	O
,	O
lead	O
to	O
the	O
natural	O
idea	O
that	O
one	O
can	O
produce	O
data	O
and	O
then	O
use	O
it	O
for	O
training	O
.	O
</s>
<s>
This	O
fully	O
synthetic	O
approach	O
has	O
not	O
yet	O
materialized	O
,	O
although	O
GANs	O
and	O
adversarial	O
training	O
in	O
general	O
are	O
already	O
successfully	O
used	O
to	O
improve	O
synthetic	B-General_Concept
data	I-General_Concept
generation	O
.	O
</s>
