1CSIRO, Perth, WA, Australia
The Australian Square Kilometer Array Pathfinder (ASKAP) is a next generation radio telescope located in Western Australia. Once fully commissioned one of its key science projects, the Evolutionary Map of the Universe (EMU), is expect to detect upwards of 70 million radio objects over five years. This absurd data volume requires new approaches to classify complex objects to extract the maximum amount of scientific knowledge. We investigate how rotationally invariant self-organizing maps (SOM) can be used as a tool to identify the predominate morphological shapes of sources as well as a tool to transfer knowledge from labelled data-sets to unlabelled subjects. Further, by exploiting the transform information learnt by the SOM, we construct an approach that is able to identify the individual components of complex sources across multiple wavelengths without requiring large training sets with known labels. As this approach requires only source positions and no labels for training, it is ideal for EMU and similar deep, all sky radio surveys from the next generation of radio telescopes.
PostDoc working at CSIRO Astronomy and Space Science researching applications of machine learning methods for the classification of radio objects.