Back to Volume
Paper: Unsupervised Catalog Classification
Volume: 77, Astronomical Data Analysis Software and Systems IV
Page: 264
Authors: Murtagh, F.
Abstract: Automatic classification of large catalogs may be carried out for survey objectives, or to elucidate the catalog's contents in a minimally-restricted way. A cohesive mathematical framework is of benefit, since otherwise the boundaries between hand-crafted analysis tools will themselves require close monitoring at all times. We investigate the use of the Kohonen self-organizing feature map (SOFM) method for simultaneous clustering and dimensionality-reduction. We find a difficulty with the summarizing properties of this method, and propose a mathematically-coherent enhancement to the SOFM method to overcome it. This involves use of an agglomerative contiguity-constrained clustering method on the SOFM output. An application to the IRAS Point Source Catalog is presented.
Back to Volume