Galaxy Morphology Without Classification: Self Organising Maps

Abstract

We examine a general framework for visualizing datasets of high (> 2) dimensionality, and demonstrate it using the morphology of galaxies at moderate redshifts. The distributions of various populations of such galaxies are examined in a space spanned by four purely morphological parameters. Galaxy images are taken from the Hubble Space Telescope (HST) Wide Field Planetary Camera 2 (WFPC2) in the I band (F814W). Since we have little prior knowledge on how galaxies are distributed in morphology space we use an unsupervised learning method (a variant of Kohonen's Self Organizing Maps, or SOMs). This method allows the data to organize themselves onto a two-dimensional space while conserving most of the topology of the original space. It thus enables us to visualize the distribution of galaxies and study it more easily. The process is fully automated, does not rely on any kind of eyeball classification and is readily applicable to large numbers of images. We apply it to a sample of 2934 galaxies, and find that morphology correlates well with the apparent magnitude distribution and to lesser extents with color and bulge dominance. The resulting map traces a morphological sequence similar to the Hubble Sequence, albeit two dimensional. We use the SOM as a diagnostic tool, and rediscover a population of bulge-dominated galaxies with morphologies characteristic of peculiar galaxies. This is achieved without recourse to eyeball classification. We also examine the effect of noise on the resulting SOM, and conclude that down to I magnitude of 24 our results are reliable. We propose using this method as a framework into which more physical data can be incorporated when they become available. Hopefully, this will lead to a deeper understanding of galaxy evolution.