Astronomers are using AI to study the vast universe — fast
The next generation of powerful telescopes will scan millions of stars and generate massive amounts of data that astronomers will be tasked with analyzing. That's way too much data for people to sift through and model themselves — so astronomers are turning to AI to help them do it.
The bottom line: Algorithms have helped astronomers for a while, but recent advances in AI — especially image recognition and faster, more inexpensive computing power —mean the techniques can be used by more researchers. “The mode of operation has to change because there is no way we can handle those data flows,” says astronomer Derek Buzasi from Florida Gulf Coast University.
One example: An 8.4-meter-mirror telescope being built in Chile — called the Large Synoptic Survey Telescope (LSST) — will take photos of the entire Southern sky every few nights for 10 years using a car-sized digital camera. All told, it is expected to collect more than 50 million gigabytes of raw data.
What's new: There are "new methods that didn’t exist 5 to 10 years ago that either improved speed or accuracy,” says Donald Lee-Brown, a graduate student at the University of Kansas. A rough analysis on his part shows the number of astronomy papers mentioning machine learning in the abstract increased five-fold over the past five years.
How they're using it:
1) To coordinate telescopes. The large telescopes that will survey the sky will be looking for transient events — new signals or sources that "go bump in the night," says Los Alamos National Laboratory's Tom Vestrand.
- Some of these events — like gamma ray bursts, which Vestrand calls "the birth announcements of black holes" — last less than a minute. In that time, they need to be detected, classified into real or bogus events (like an airplane flying past), and the most appropriate telescopes turned on them for further investigation.
- With a telescope like the LSST, there could be on the order of 50,000 transient events each night and hundreds of telescopes around the world working in concert.
- "This isn't a place to have people typing fast," says Vestrand. "It has to be machine to machine."
2) To analyze data. Every 30 minutes for two years, NASA’s new Transiting Exoplanet Survey Satellite will send back full frame photos of almost half the sky, giving astronomers some 20 million stars to analyze.
- “There will be more data on these stars than we’ve ever seen in total before,” says Buzasi. The idea then is that an AI can classify it, lump together the weird ones if they have some similarities, and hand them off to humans to "look at the 1% the AI can’t figure out.”
- “A tool that uses neural networks can get temperatures or metallicities [of stars] that are not only more precise estimates than our old methods of measuring these features, but we get it billions of times faster" than a graduate student could, says Lee-Brown.
- Machine learning is being used to study black holes, find exoplanets and model the universe and its parameters.
- Buzasi says AI can help most with consistency in treating data the same way which is hard to do when humans are involved.
3) To mine data. "Most astronomy data is thrown away but some can hold deep physical information that we don’t know how to extract," says Joshua Peek from the Space Telescope Science Institute.
- After those beautiful images of nebulae are produced, the information in them is often discarded, says Peek. He is developing convolutional neural networks, a machine learning tool for classifying images into different kinds of objects, to extract information about features in these structures of diffuse plasma and gas — the state in which most of the universe's normal matter exists. Astronomers could then look at what's common or not among different structures in the universe.
One big question: "How do you write software to discover things that you don’t know how to describe?" asks Vestrand. "There are normal unusual events, but what about the ones we don’t even know about? How do you handle those? That will be where real discoveries happen because by definition you don’t know what they are."