Computing the distance between point a and point b is typically considered to be very easy. However, there are times when computing a distance can take significant computation time; we call these expensive distance metrics. Suppose we have some expensive distance metric and we need to compute the distances between a bunch of points. This paper explores a method that to reduce the number of queries to the distance metric and the effect on clustering. The authors find that total run time can be reduced while only inducing small inaccuracies in clustering output.

Computing the distance between point a and point b is typically considered to be very easy. However, there are times when computing a distance can take significant computation time; we call these expensive distance metrics. Suppose we have some expensive distance metric and we need to compute the distances between a bunch of points. This paper explores a method that to reduce the number of queries to the distance metric and the effect on clustering. The authors find that total run time can be reduced while only inducing small inaccuracies in clustering output.