Machine Learning Techniques for the Square Kilometre Array
Classifying the vast numbers of radio sources detected by the Square Kilometre Array (SKA) and its pathfinders will be impossible to do with current techniques. Traditionally a radio source is examined by hand, classified and a spectroscopic redshift obtained. This will not be possible with the vast number of faint and distant radio sources. The Evolution Map of the Universe (EMU) project on the Australian SKA Pathfinder is expected to detected 70 million radio sources alone.
This project will focus on using machine learning techniques and other advanced algorithms for the three stages of radio source identification:
- Association: matching possibly multiple radio sources to one host galaxy in a robust and automatic fashion.
- Estimation: what is the distance to the radio source? can we improve on the classic photometric redshift technique.
- Classification: what is the morphology of the radio source? Is it powered by a black hole or star formation.
This project will examine a series of machine learning methods to obtain redshifts and identifications based on limited information including kth Nearest Neighbour and Self-Organised Map amongst others. These will be tested against a growing catalogue of known radio sources in particular those obtained from the OzDES project. Hence, this project will also be actively involved in a major observational programme on the Anglo-Australian Telescope along with reduction and classification of optical spectra.
Once these methods are well calibrated then even sources with large uncertainties on their distances can be used among the millions of radio sources from MWA and ASKAP to constrain the evolution of the star forming and black hole radio populations. This work will be of key importance in the preparation for deep surveys with the Square Kilometre Array.