Proof of Theorem 1. Let be the nearest neighbor
of the query
in the training set
. We
show that as long as we independently draw the training samples,
the probability that
traverses the recursive partition tree
exactly same as
approaches 1 as
. Assume
at the subregion
, we have the projection matrices V and
W. The mixture distance from
to a center
C under region
is
Using the triangle inequality property of the mixture distance, we have
If the sum of the last three terms ( )
of the above equation
, we would have
which means that the approximator would
make the same decision for
and
. Let
be the
event that the nearest neighbor of
in the training set
has a distance larger than
while
denotes
the event that the the nearest neighbor of
in the single
has a distance larger than
. Since each
is independently
and identically drawn, we have
On the other hand, is a positively supported, which
means that
. Thus,
which approaches zero when . So, given any value
and
, we can have a positive N, so that
for any n>N, we have
Now, we have shown that the approximator makes
the same decision for and
as
.
Finally, the fact that
is an interior point guarantees that
)=f(
as
.