Exoplanet A.I.

Artist’s rendition of the Kepler Space Telescope. Credit: NASA, Wendy Stenzel, Daniel Rutter.

Until NASA retired it in 2018, the Kepler Space Telescope sent all the interesting data it could collect about stars from space to Earth. Which of these points were useful and which weren’t would get decided later, as information from its unique vantage point was recorded.

Even now, so much data remains from its nine years in orbit that researchers are still making discoveries from the archives it left behind.

That includes researchers like Anne Dattilo, who was an astronomy undergraduate at The University of Texas at Austin when, with the help of artificial intelligence, she discovered two new planets hiding in Kepler’s K2 mission data.lanets.

A member of Women in Natural Sciences and an alumna of the Freshman Research Initiative, Dattilo used the A.I. called AstroNet-K2 to find planets. She fed it data on known examples, a process known as “supervised learning.” Then AstroNet-K2 found two new examples that fit a similar profile, previously unknown planets outside our solar system, or exoplanets, unlovingly labeled K2-293b and K2-294b. The pair were later confirmed by ground-based telescope observations.

Creating an automatic, unbiased system to identify signals of interest— in this case, a temporary dimming of starlight that occurs when a planet orbits its star — is only the beginning, according to Dattilo, who is now in graduate school.

“There may be unusual planetary systems out there that are interesting but would not be recognized by our model,” she observed. This means the A.I. method, if applied to other data sets or telescopes, “could potentially find highly valuable, previously missed planets.”