Exoplanet K2-18b (Artist’s Impression)
This artist’s impression shows the planet K2-18b, its host star and an accompanying planet. Its discovery came from "citizen scientists" examining publicly stored data, but one machine learning expert said this type of scanning could be done by computers in the future. (ESA/Hubble, M. Kornmesser)

Computers could give astronomers a badly needed helping hand

In Sky News This Week: Machine learning has been around for decades, but it will enter likely the mainstream in astronomy soon.

Astronomy has a looming problem — a problem of data overwhelm.

All-sky survey telescopes, exoplanet hunters and forthcoming observatories like the James Webb Space Telescope search out space with the hope of catching interesting things in the sky.

Then scientists must parse all that information and come to conclusions, using all the old information flowing in while keeping an eye on older data that may be useful. If university students are facing paper research overwhelm with all the databases available, it’s even worse for those few professional astronomers seeking the cutting edge of their field.

For example: a NASA telescope called Roman — short for the Nancy Grace Roman Space Telescope — will fly into space in the 2020s for wide-field sky observing. But the datasets it will send down could have millions of galaxies in a single shot, no surprise as each photograph could have a field of view 100 times wider than the Hubble Space Telescope. It’s too much for even a large astronomy team to parse, according to a recent Space Telescope Science Institute release (STScI being the entity that manages both Hubble and Roman).

Enter the solution: the quickly growing field of machine learning, which allows computers to scan immense datasets, get training on which patterns are important, then identify patterns themselves to present the data to researchers. Machine learning (a facet of artificial intelligence, or AI) has been around for decades, but with quantum computers and cloud storage quickly maturing, it will enter the mainstream in astronomy very shortly.

AI is not a panacea. Machine learning has a facet of taking place in the dark, in that the computer can make choices that the humans can’t easily replicate — which is bad for scientific integrity if not properly controlled. Researchers also caution that using AI is not the ultimate solution to data overwhelm, and that it is too easy to throw it around as a “buzzword” to get government funding or media attention. But used properly, AI could give astronomers a very precious resource: time.

“It gets increasingly hard to process these data manually. We need automatic methods to take over some of this,” Bernhard Schölkopf, director of the Max Planck Institute for Intelligent Systems in Germany and a fellow at the Canadian Institute for Advanced Research who focuses on learning in machines and brains, in an interview.

Schölkopf pointed to the 2015 discovery of exoplanet K2-18b, a “super-Earth” orbiting a red dwarf star, as one example of machine learning’s potential. Data sometimes needs to be parsed multiple times to find things, he said, pointing to crowdsourcing astronomy photographs as an example.

While the discovery of K2-18b came from “citizen scientists” using simple scanning methods to examine publicly stored data, before long, computers will take over some of the easier functions. AI can also control telescopes automatically to find, for example, the most interesting events in the sky on a particular night, Schölkopf said.

In a few words, using AI will save time for astronomers and allow them to focus their expertise less on scanning databases and more on what they’re good at: analysis. While machine learning and AI may still be somewhat cutting edge, they will soon become part of the “standard toolkit” for astronomers, Schölkopf said — much like how computerized data analysis is now becoming an integral part of investigative journalism, which also relies on vast databases of information.

For Roman, Canadians are keeping a close eye on the forthcoming AI opportunity – not least because Canuck companies ABB and Nuvu are both supplying camera technology that will assist in taking those epic pictures.

“Roman’s primary mission is a survey,” said Frederic J. Grandmont, ABB’s technology and business development manager, in an interview, adding that we start to talk about AI in astronomy (and other fields) when the processing improves as more data comes in.

“Survey data pipeline are inherently oriented toward applying a known algorithm to a massive dataset,” he added. “In many cases, astronomers will get more if they orient their processing toward some level of [computer] autonomy, because there is simply too much data to look at on a computer screen.”

The looming problem of integrating older data is also on the mind of astronomers, Grandmont said. The Canadian Astronomy Data Center is one of several international entities trying to put the information available into the right formats for computers to use.

“We need AI to filter and sort all this data, a bit like Google manages to classify so many web pages to allow one individual to pull the most relevant records with a few key words in less than a second,” Grandmont said. “That does not exist yet in astronomy, and it holds promise for more hidden discoveries.”

Sky News This Week is a biweekly column by Canadian science and space journalist Elizabeth Howell, focusing on a trending news topic in Canadian astronomy and space.