<s>
An	O
Inference	B-Algorithm
Attack	I-Algorithm
is	O
a	O
data	B-Application
mining	I-Application
technique	O
performed	O
by	O
analyzing	O
data	O
in	O
order	O
to	O
illegitimately	O
gain	O
knowledge	O
about	O
a	O
subject	O
or	O
database	O
.	O
</s>
<s>
An	O
Inference	B-Algorithm
attack	I-Algorithm
occurs	O
when	O
a	O
user	O
is	O
able	O
to	O
infer	O
from	O
trivial	O
information	O
more	O
robust	O
information	O
about	O
a	O
database	O
without	O
directly	O
accessing	O
it	O
.	O
</s>
<s>
The	O
object	O
of	O
Inference	B-Algorithm
attacks	I-Algorithm
is	O
to	O
piece	O
together	O
information	O
at	O
one	O
security	O
level	O
to	O
determine	O
a	O
fact	O
that	O
should	O
be	O
protected	O
at	O
a	O
higher	O
security	O
level	O
.	O
</s>
<s>
While	O
inference	B-Algorithm
attacks	I-Algorithm
were	O
originally	O
discovered	O
as	O
a	O
threat	O
in	O
statistical	O
databases	O
,	O
today	O
they	O
also	O
pose	O
a	O
major	O
privacy	O
threat	O
in	O
the	O
domain	O
of	O
mobile	B-Application
and	O
IoT	B-Operating_System
sensor	O
data	O
.	O
</s>
<s>
Data	O
from	O
accelerometers	O
,	O
which	O
can	O
be	O
accessed	O
by	O
third-party	O
apps	O
without	O
user	O
permission	O
in	O
many	O
mobile	B-Application
devices	I-Application
,	O
has	O
been	O
used	O
to	O
infer	O
rich	O
information	O
about	O
users	O
based	O
on	O
the	O
recorded	O
motion	O
patterns	O
(	O
e.g.	O
,	O
driving	O
behavior	O
,	O
level	O
of	O
intoxication	O
,	O
age	O
,	O
gender	O
,	O
touchscreen	O
inputs	O
,	O
geographic	O
location	O
)	O
.	O
</s>
<s>
Highly	O
sensitive	O
inferences	O
can	O
also	O
be	O
derived	O
,	O
for	O
example	O
,	O
from	O
eye	B-General_Concept
tracking	I-General_Concept
data	O
,	O
smart	B-Device
meter	I-Device
data	O
and	O
voice	O
recordings	O
(	O
e.g.	O
,	O
smart	B-Device
speaker	I-Device
voice	O
commands	O
)	O
.	O
</s>
