<s>
In	O
pattern	O
recognition	O
,	O
the	O
iDistance	B-Algorithm
is	O
an	O
indexing	O
and	O
query	O
processing	O
technique	O
for	O
k-nearest	B-General_Concept
neighbor	I-General_Concept
queries	I-General_Concept
on	O
point	O
data	O
in	O
multi-dimensional	O
metric	O
spaces	O
.	O
</s>
<s>
The	O
kNN	O
query	O
is	O
one	O
of	O
the	O
hardest	O
problems	O
on	O
multi-dimensional	O
data	O
,	O
especially	O
when	O
the	B-Algorithm
dimensionality	I-Algorithm
of	I-Algorithm
the	I-Algorithm
data	I-Algorithm
is	I-Algorithm
high	I-Algorithm
.	O
</s>
<s>
The	O
iDistance	B-Algorithm
is	O
designed	O
to	O
process	O
kNN	O
queries	O
in	O
high-dimensional	O
spaces	O
efficiently	O
and	O
it	O
is	O
especially	O
good	O
for	O
skewed	B-General_Concept
data	I-General_Concept
distributions	I-General_Concept
,	O
which	O
usually	O
occur	O
in	O
real-life	O
data	O
sets	O
.	O
</s>
<s>
The	O
iDistance	B-Algorithm
can	O
be	O
augmented	O
with	O
machine	O
learning	O
models	O
to	O
learn	O
the	O
data	O
distributions	O
for	O
searching	O
and	O
storing	O
the	O
multi-dimensional	O
data	O
.	O
</s>
<s>
Building	O
the	O
iDistance	B-Algorithm
index	O
has	O
two	O
steps	O
:	O
</s>
<s>
Using	O
cluster	B-Algorithm
centers	I-Algorithm
as	O
reference	O
points	O
is	O
the	O
most	O
efficient	O
way	O
.	O
</s>
<s>
Succinctly	O
,	O
the	O
data	O
points	O
are	O
partitioned	O
into	O
Voronoi	B-Architecture
cells	I-Architecture
based	O
on	O
well-chosen	O
reference	O
points	O
.	O
</s>
<s>
This	O
distance	O
plus	O
a	O
scaling	O
value	O
is	O
called	O
the	O
point	O
's	O
iDistance	B-Algorithm
.	O
</s>
<s>
By	O
this	O
means	O
,	O
points	O
in	O
a	O
multi-dimensional	O
space	O
are	O
mapped	O
to	O
one-dimensional	O
values	O
,	O
and	O
then	O
a	O
B+	O
-tree	O
can	O
be	O
adopted	O
to	O
index	O
the	O
points	O
using	O
the	O
iDistance	B-Algorithm
as	O
the	O
key	O
.	O
</s>
<s>
The	B-Algorithm
iDistance	I-Algorithm
technique	I-Algorithm
can	O
be	O
viewed	O
as	O
a	O
way	O
of	O
accelerating	O
the	O
sequential	O
scan	O
.	O
</s>
<s>
Instead	O
of	O
scanning	O
records	O
from	O
the	O
beginning	O
to	O
the	O
end	O
of	O
the	O
data	O
file	O
,	O
the	O
iDistance	B-Algorithm
starts	O
the	O
scan	O
from	O
spots	O
where	O
the	O
nearest	B-General_Concept
neighbors	I-General_Concept
can	O
be	O
obtained	O
early	O
with	O
a	O
very	O
high	O
probability	O
.	O
</s>
<s>
The	O
iDistance	B-Algorithm
was	O
first	O
proposed	O
by	O
Cui	O
Yu	O
,	O
Beng	O
Chin	O
Ooi	O
,	O
Kian-Lee	O
Tan	O
and	O
H	O
.	O
V	O
.	O
Jagadish	O
in	O
2001	O
.	O
</s>
