The bright object in the centre of this Hubble Space Telescope image is galaxy cluster Abell 383. It is used as a gravitational lens to view the distant galaxies in the background. The effects can be seen in the twisted and warped light surrounding the cluster. Image Credit: NASA, ESA, J. Richard (CRAL) and J.-P. Kneib (LAM). Acknowledgement: Marc Postman (STScI)
Artificial intelligence is being trained to help astronomers search for gravitational lenses, the warping effect of large cosmic masses on light from distant objects.
If fine-tuned, the programme could help scientists learn more about dark matter.
AI is becoming increasingly common in our everyday lives, from social media websites that customise our news feeds, to the self-driving cars that may one day become commonplace.
AI can also be used by astronomers to sift through reams of data on our Universe that is continually being collected by telescopes.
Astronomers from universities in Groningen, Naples and Bonn are currently fine-tuning AI to help search for gravitational lenses.
Gravitational lensing refers to how massive objects in space can actually bend and warp the light from objects that are farther away.
The theory was predicted by Albert Einstein, and enables astronomers to use huge objects like galaxy clusters as cosmic magnifying glasses, enabling them to see objects that would otherwise be too far away to observe clearly.
The hunt for gravitational lenses involves searching through thousands of images captured by telescopes, in order to single out those that show gravitational lensing taking place.
This is done because the results can aid in the search for dark matter; an allusive, invisible substance that, together with dark energy, is thought to make up the majority of the Universe.
So-called ‘convolutional neural networks’ can help with this.
With the help of artificial intelligence, astronomers discovered 56 new gravity lens candidates. This picture shows a sample of the handmade photos of gravitational lenses that the astronomers used to train their neural network. Image Credit: Enrico Petrillo (Rijksuniversiteit Groningen)
The team of astronomers have been training a neural network using millions of homemade images of gravitational lenses, enabling it to ‘learn’ what to look for.
They then showed the network millions of images from a patch of the sky 255 square degrees across; just over half a per cent of the sky.
The neural network cut these images down to 761 gravitational lens candidates, and the astronomers were then able to further reduce this number down to 56, which will be confirmed by follow-up investigations with telescopes like the Hubble Space Telescope.
The neural network also rediscovered two known lenses, but missed a third known lens.
This third lens was smaller, and the neural network had not been trained to spot that size yet.
“This is the first time a convolutional neural network has been used to find peculiar objects in an astronomical survey,” says Carlo Enrico Petrillo of the University of Groningen in the Netherlands and first author of the study.
“I think it will become the norm since future astronomical surveys will produce an enormous quantity of data which will be necessary to inspect.
We don’t have enough astronomers to cope with this.”