Discussion:

Similarity based clustering has useful applications in providing ``natural'' and unsupervised classification schemas for our data. Other algorithms help us identifying representative points within the data, i.e. the K-center algorithm. In general, clustering methods are mainly based on the distance measure among the variables. Mathematical algorithms use these distances to group data entities into similarity or representative clusters. Therefore there is no direct linkage to most basic techniques, but many similarities can be observed: the field selection for input to the metric is projection; calculating the distances is creating a new meta-dimension; each cluster is a subset; the induced partition is coarsening; and the derived labels on the clusters can act as a new dimension. For ``supervised clustering'' projection and subsetting is used to relate the obtained clusters to the supervised classification. Furthermore many other methods can be applied and use this new dimension.