Wednesday

Morphology encodes key details of the assembly histories of galaxies. While past work on morphological classification includes both algorithmic techniques and visual methods, the sheer size of forthcoming datasets (e.g. LSST, EUCLID etc.) makes it critical to develop accurate automated methods that are guided by visual classification (e.g. via Galaxy Zoo). We present a technique based on unsupervised machine learning that is ideal for analysing current and future observational surveys. Distinct from previous approaches, this technique requires no pre-labelled training set. Instead it automatically identifies objects that are similar in shape and colour and offers the capability to perform a galaxy similarity search. We demonstrate the technique on the five HST CANDELS fields and compare our machine-based classifications to human-classifications from the Galaxy Zoo: CANDELS project. We find that the technique not only cleanly separates major morphological classes (e.g. spirals and ellipticals), but also offers an efficient route to identifying morphological details such as colour and the presence of interactions. We discuss the future potential of such technique in the context of forthcoming surveys like EUCLID and LSST (for which we are building a morphological pipeline using this algorithm). A morphological catalogue for the CANDELS fields is available at www.galaxyml.uk