Machine Learning Finds 301 More Planets in Kepler Data

IMAGE: Over 4,500 planets have been discovered round different stars, however scientists anticipate that our galaxy accommodates thousands and thousands of planets. There are a number of strategies for detecting these small, faint our bodies round a lot bigger, shiny stars. CREDIT: NASA/JPL-Caltech By Beth Johnson

Not to be outdone by TESS, the Kepler mission continues to develop our exoplanet catalog by leaps and bounds. In a brand new paper in The Astrophysical Journal, scientists used a deep neural community known as ExoMiner and NASA’s Supercomputer Pleiades to comb via Kepler and K2 mission knowledge in an try to hurry up the method of discovering and confirming exoplanets.

Kepler and TESS each use the identical methodology of detecting exoplanets — the transit methodology — the place we measure the quantity of sunshine coming from a star and see if there are any dips in that mild that might point out the presence of an orbiting planet. There are a couple of points with this methodology, and considered one of them is that false positives can seem as a consequence of one other star or starspots and even mud. And taking a look at every star with a human eye is extremely time-consuming. I do know. I’ve finished it.

And that’s the place machine studying comes in. We can prepare a neural community to distinguish between an exoplanet and people false positives utilizing knowledge we’ve got already processed. And with over 4,500 exoplanets in the Kepler knowledge already validated, there may be loads of knowledge to make use of for coaching. So that’s what programmers did in creating ExoMiner, and it appears to have labored brilliantly. As undertaking lead Hamed Valizadegan mentioned: ExoMiner is very correct and in some methods extra dependable than each current machine classifiers and the human specialists it’s meant to emulate due to the biases that include human labeling.

While none of those 301 new exoplanet discoveries are considered Earth-like or in their liveable zones, they do add to the rising mounds of knowledge we are able to use for evaluating exoplanet populations and techniques. Congratulations to the Kepler and ExoMiner groups, together with SETI Institute scientists Doug Caldwell and Jeff Smith. We stay up for seeing what different discoveries you make.

More InformationNASA JPL press releaseUSRA press launch

“ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier to Mine Exoplanets,” Hamed Valizadegan et al., to be revealed in The Astrophysical Journal (preprint on

This story was written for the Daily Space podcast/YouTube sequence. Want extra information from myself, Dr. Pamela Gay, and Erik Madaus? Check out article was initially revealed on




Recommended For You