Humans can be irreplaceable when it comes to discovering new celestial bodies thanks to our pattern detection skills. But sometimes computers can spot things that humans can’t, including the recent discovery of exoplanets using machine learning.
The exoplanet was discovered in a protoplanetary disk called HD 142666 by researchers at the University of Georgia. A protoplanetary disk is a rotating disk of gas swirling around a young star, from which planets form. Planets form within these disks as matter clumps together until eventually enough gravity pulls in more matter. The researchers examined a series of previous observations across the protoplanetary disk and used machine learning models to search for exoplanets. I may have missed it the first time. They identified one disk where planets are likely, based on the unusual way gas moves within the disk.
“We used traditional methods to identify the planets, but our model told us to run those simulations and show us exactly where the planets are,” said the lead author. “When we applied the model to a set of older observations, we identified a disk that had already been analyzed but was not known to host planets,” said Jason Terry of ‘India’ in a statement. We ran a simulation of the disk and found that the planet could replicate our observations, similar to our findings.”
The researchers say this is a proof-of-concept that machine learning can be used to make new discoveries of exoplanets, even on previously analyzed data. This could mean that more exoplanets will be discovered in the future, and it is also possible that discoveries will be made more quickly.
“This shows that our models (and machine learning in general) have the ability to quickly and accurately identify important information that people might miss. and subsequent theoretical insights can be dramatically sped up,” said Terry. “It only took about an hour to analyze that entire catalog and find strong evidence that there is a new planet at a particular location. I think there is a place for that.”
The study is published in The Astrophysical Journal.
Editor’s pick
