Distributed Data Clustering

To make effective use of distributed information, it is desirable to allow coordination and collaboration among various information sources. This paper deals with clustering data emanating from different sites. The process of clustering consists of three steps: find the (local) clusters of data at each site; find (higher) clusters from the union of the distributed data sets at the central site; and finally compute the associations between the two sets of clusters. The approach aims at discovering the hidden structure of a multi-source data and assigning unseen data points coming from a site to the right higher cluster without any need to access their feature values. The proposed approach is evaluated experimentally.