Facebook facial recognition AI now spots ‘burping’ supermassive black holes

31 Oct 2018

Centaurus A is our nearest giant radio galaxy, at a distance of about 13m light years away in the southern constellation of Centaurus. Image: ESO/IDA/Danish 1.5 m/R Gendler, J-E Ovaldsen and S Guisard

AI once used to spot faces on Facebook has been adapted to identify radio galaxies in the farthest parts of the known universe.

When it comes to discussing what the future of astronomy will be, it would be impossible to ignore the role artificial intelligence (AI) will play in locating and cataloguing billions of galaxies significantly faster than humans ever could.

One such new AI – called ClaRAN – has been revealed by researchers from the University of Western Australia node of the International Centre for Radio Astronomy Research. The centre’s job is to spot the powerful radio jets ‘burped’ by supermassive black holes at the centre of radio galaxies.

ClaRAN grew out of an open source version of Microsoft and Facebook’s object detection software and, in the latter’s case, faces. Now, in a paper published to the Monthly Notices of the Royal Astronomical Society, the researchers detailed how they completely overhauled this AI to train it to recognise galaxies instead of people.

14 blue panels showing predicted radio galaxies in pink.

14 radio galaxy predictions ClaRAN made during its scan of radio and infrared data. All predictions were made with a high ‘confidence’ level, shown as the number above the detection box. Image: Dr Chen Wu and Dr Ivy Wong, ICRAR/UWA

‘This is the future of programming’

The open source AI will be paired with the Australian Square Kilometre Array Pathfinder telescope and it is expected to observe up to 70m galaxies.

While traditional algorithms can identify up to 90pc of radio galaxy sources correctly, there remains 10pc – equivalent to 7m galaxies – that can be considered ‘difficult’ and require a human to look over them.

“If ClaRAN reduces the number of sources that require visual classification down to 1pc, this means more time for our citizen scientists to spend looking at new types of galaxies,” said Dr Ivy Wong of the team.

Her colleague, Dr Chen Wu, added: “All you do is set up a huge neural network, give [ClaRAN] a ton of data and let it figure out how to adjust its internal connections in order to generate the expected outcome.

“The new generation of programmers spend 99pc of their time crafting the best quality datasets and then train the AI algorithms to optimise the rest. This is the future of programming.”

Colm Gorey was a senior journalist with Silicon Republic

editorial@siliconrepublic.com