Large scale classification of celestial objects

Research areas

Temporary Supervisor

Dr Cheng Soon Ong


The ANU SkyMapper telescope is surveying the Southern sky in the optical spectrum across 0.35 to 1 micron wavelength. Millions of celestial objects are found with different colours, shapes and sizes. This huge resource allows us to study the structure and evolution of the Milky Way as well as that of millions of other galaxies. This project aims at training a computer to classify them automatically into meaningful categories and sort them by physical properties. Will be co-supervised by Christian Wolf from RSAA.


Traditional methods for multi-class classification rely on the closed world assumption, that is the assumption that the set of possible classes is fixed and known during training. However in many applications, including classification of astronomical objects, it may be difficult to obtain a catalog and annotations for all possible classes of objects [1]. In recent years, there have been several proposals for systems that simultaneously identify the class of the object and detect whether the object comes from a novel class [2]. These systems have been applied to object recognition in the computer vision domain, and can be analysed under the class contamination setting [3,4]. The student would work closely with experts in machine learning and astrophysics to develop and build an open world multi-class classification tool for celestial objects.


An ideal candidate would have a background in computational sciences and would be comfortable with implementation of numerical algorithms. Programming skills in Python or C/C++, and an understanding of machine learning tools would be necessary for the empirical parts. The project provides an opportunity for the student to apply machine learning techniques to a cutting edge scientific discovery problem.

Background Literature

[1] J. Denzler, E. Rodner, P. Bodesheim, A. Freytag Beyond the closed-world assumption: The importance of novelty detection and open set recognition Unsolved Problems in Pattern Recognition and Computer Vision (GCPR Workshop) 2013. [2] D.M.J. Tax and R.P.W. Duin Growing a multi-class classifier with a reject option Pattern Recognition Letters, vol.29, pp.1565-1570, 2008 [3] T. Sanderson and C. Scott Class Proportion Estimation with Application to Multiclass Anomaly Rejection AISTATS 2014 [4] A. Menon, B. van Rooyen, C.S. Ong, R.C. Williamson Learning from Corrupted Binary Labels via Class-Probability Estimation ICML 2015


To develop and build state of the art machine learning methods on an exciting big data problem.

Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing