Fifty potential planets were confirmed with help from a new machine-learning algorithm developed via scientists at the University of Warwick, according to a new study published in the Monthly Notices of the Royal Astronomical Society.
Machine learning confirms 50 new planets
Astronomers used a process based on machine learning (a type of artificial intelligence) to analyze a sample of potential planets and discern which were real or "fake," or false positives — for the first time.
The team's results were reported in the new study, wherein they also performed the first large-scale compare-and-contrast of novel planet validation techniques. These include the newly-applied machine learning algorithm, which will see use to statistically confirm future exoplanet discoveries.
Typically, exoplanet surveys search massive quantities of data gathered via telescopes for signs of planets passing between Earth and their host star — in a process called transiting. When it happens, the star's light dips in intensity to a degree telescopes pick up, but the dips can also happen in binary star systems, background interference, or even camera errors. Taken together, these potential sources of interference call for a means of distinguishing real from "fake" exoplanet indications.
Training machine learning to search for exoplanets
This is why researchers from Warwick's departments of physics and computer science, in addition to the Alan Turing Institute, built a machine learning-based algorithm capable of differentiating real planets from fake ones in large, thousand-candidate samples identified during telescope missions like NASA's TESS and Kepler, according to phys.org.
The machine learning-method was trained to correctly identify real planets with help from two large samples of confirmed planets and false positives from the now-defunct Kepler mission. Then the researchers employed the algorithm on a new dataset of erstwhile-unconfirmed planetary candidates gathered via Kepler. The results unveiled 50 new confirmed planets — the first validation from machine learning.
Earlier machine learning techniques capably ranked planet candidates, but were never able to distinguish the probability that a candidate was in fact a planet without help — which is the main purpose for planet validation.
The 50 new planets range in type from the size of Neptune to the exciting potential of Earth-like scales, with orbits up to 200 days and as low as one, single day. Now possessed of the knowledge that the 50 planet candidates are not fakes, astronomers may move forward with ongoing observations of the newly-discovered exoplanets via committed telescopes.
Machine learning will accelerate exoplanet validation
Professor David Armstrong of the University of Warwick's department of physics said: "The algorithm we have developed lets us take 50 candidates across the threshold for planet validation, upgrading them to real planets. We hope to apply this technique to large samples of candidates from current and future missions like TESS and PLATO. In terms of planet validation, no-one has used a machine learning technique before."
"Machine learning has been used for ranking planetary candidates but never in a probabilistic framework, which is what you need to truly validate a planet," he added. "Rather than saying which candidates are more likely to be planets, we can now say what the precise statistical likelihood is. Where less than a 1% chance of a candidate being a false positive, it is considered a validated planet."
As a new suite of space-based telescopes begins missions to seek out new worlds possibly hosting new civilizations, we can be sure that many if not most of the planets confirmed to be more than errant cosmic noise will get their validation from machine learning.