Training machines to expect the unexpected

Interesting project.

The case study of the Nobel-prize-winning discovery of pulsars by Jocelyn Bell is instructive. A talented and persistent PhD student studying interstellar scintillation (and thus expanding the observational phase space), and who knew her instrument intimately, recognised that ‘bits of scruff’ on the chart recorder could not be terrestrial interference, but represented a new type of astronomical object (Bell Burnell 2009). As a result, she discovered pulsars.

...A present-day Jocelyn Bell is unlikely to understand the instrument well enough to distinguish astrophysical phenomena from instrumental effects, and would not be able to sift through the petabytes by hand, searching for something unusual. On the other hand, failure to identify unexpected effects may mean missing out on the most important science to emerge from ASKAP. It is therefore necessary to plan explicitly to build techniques to make unexpected discoveries, rather than hoping to stumble across them.

We have therefore started a project called “WTF”, which explicitly aims to mine EMU data to discover unexpected science that is not part of our primary science goals, using a variety of machine-learning techniques and algorithms. Although targeted specifically at EMU, we expect this approach to have broad applicability to astronomical survey data.