How artificial intelligence helps astronomers in the era of ‘big data’ - IPM Newsroom

URBANA — We are in what many astronomers are calling the era of “big data” in astronomy. As observation technology advances, the amount of data collected grows, resulting in data sets that are astronomical.

The Legacy Survey of Space and Time is one example. The survey will collect data over the next ten years, contributing to a variety of research topics, including dark energy and how our galaxy formed.

“That’s expected to observe billions of objects — basically look at the entire southern sky,” said Grant Merz, an astronomy graduate student at the University of Illinois Urbana-Champaign. “It’ll be sort of this unprecedented amount of optical imaging data.”

According to the LSST mission statement, the survey will collect over 20 terabytes of data every night — roughly equivalent to 6,000 high-definition movies’ worth of data. As exciting as that is, Merz said, going through all of that data is time-consuming and inefficient.

But by utilizing artificial intelligence, areas of interest in the data can be found quickly.

“It would be like having a daycare where you’re looking after one kid, and all of a sudden there’s 100 kids that you have to look out for,” said Amanda Wasserman, another U of I astronomy graduate student. “You’re like, ‘Oh my god, there’s no way I can look out for all 100. I need a robot to come in and help me look after them.’”

Wasserman is part of a team preparing for the flood of data the LSST will bring. She works on an AI tool called the Recommendation System for Spectroscopic Follow-up, or RESSPECT, which is trained to pick out which data in the plethora is most worth a closer look by researchers. 

For Wasserman’s research, RESSPECT can pick out which supernovae should be looked at with spectroscopy to yield information about the elements present in the supernova, she said.

Wasserman said one of her colleagues worked on another program called LAISS, Lightcurve Anomaly Identification and Similarity Search. It uses an algorithm developed for Spotify to find similar songs to recommend to users. Instead of song data, LAISS compares astronomical data.

Wasserman said LAISS can also be used to find data that is notsimilar; it can look through thousands of objects and find the anomalous ones, which astronomers likely would want to look at.

“Even just to decide if something is a supernova or not, people are combing through the data, and there are grad students and professors and postdocs spending an hour or two a day just looking at the data,” Wasserman said. “So if there’s something that is able to come in and do that for you, you can actually focus on the science that you want to do.”

There are concerns about AI’s environmental impact, but Merz said he believes the astronomy community is being careful.

“Oftentimes you’ll see in papers, people will sort of make an estimate in the amount of carbon used in producing and implementing their models,” Merz said. “I think, as a community, astronomy is very mindful of that and is generally not trying to just build the biggest and fastest model at the expense of a lot of energy.”