Machine Learning for Source Matching
A major function of the Virtual Observatory will be the ability to match up data on the same objects from disparate archive sources. A classic example is the task of finding optical counterparts to radio sources. Traditionally this has been done with simple statistical methods.
We have commenced a pilot project at the University of Queensland in collaboration with colleagues from computer science to apply new machine learning techniques to this problem. We will investigate new algorithms for automated classification such as Support Vector Machines. The radio sources are from theHIPASS catalogue and the potential optical counterparts are from the SuperCOSMOS Sky Surveys (SSS).
Further details of our project can be found at http://www.physics.uq.edu.au/people/mjd/kernel.