AI Discovers 50 new Planets Using Old Dataset from NASA

NASA‘s planet-hunting telescopes are more advanced than ever before.

In the last few decades, they’ve located several thousand exoplanets or planets outside our solar system. Of course, that also means scientists have to sift through tons of data to separate the real planets from space rocks.

Unfortunately, current techniques for confirming the existence of other planets are far from perfect. It is subject to factors such as noise, interference of an object in the background, and camera errors.

For example, telescopes such as NASA’s Transiting Exoplanet Survey Satellite (TESS) look for a dip in brightness.

Ideally, the change in brightness suggests that a planet is passing by a star. But that’s not always the case.

Sometimes, the passerby could be dust, an asteroid, or a quirk of a binary star system. It could even be the result of a glitch in the system.

Now, researchers at the University of Warwick have designed a system to separate the false positives from the real planets.

Training a Planet-Hunting Algorithm Using NASA Dataset

The researchers created a machine learning algorithm and trained it using NASA data. These include two samples of the confirmed planet and false positives from the retired Kepler mission.

After that, the researchers then used the algorithm to analyze a dataset of unconfirmed planet candidates — also from Kepler data. The model calculated the probability of each candidate being a planet.

The AI was able to confirm 50 planets, ranging from one that’s as big as Neptune to another that’s as small as Earth.

In a statement about the project, lead author and researcher from the University of Warwick, David Armstrong said:

“Rather than saying which candidates are more likely to be planets, we can now say what the precise statistical likelihood is. Where there is less than a 1 percent chance of a candidate being a false positive, it is considered a validated planet.”

According to the team, it’s the first time an algorithm validates the existence of a potential new planet.

The researchers also noted that the system is faster than existing techniques. What’s more, it can complete the planet verification process automatically.

With that said, more training is necessary for the researchers to apply the algorithm in future candidates.

“We hope to apply this technique to large samples of candidates from current and future missions like TESS and PLATO,” Armstrong added.