DIMACS TR: 2006-20
Contour approximation of data and the harmonic mean
Author: Marina Arav
A contour approximation of data is a function capturing the data
points in its lower level sets. Desirable properties of contour
approximation are posited, and shown to be satisfied uniquely (up
to a multiplicative constant) by the weighted harmonic mean of
distances to the cluster centers. This harmonic mean is the joint
distance function used in probabilistic clustering, expressing the
uncertainty of classification.
Keywords: Harmonic mean, quasi-linear mean, clustering, contour
approximation of data, distance functions.
Paper Available at:
DIMACS Home Page