Selections from 2018: Identifying Exoplanets with Deep Learning

1

Editor’s note: In these last two weeks of 2018, we’ll be looking at a few selections that we haven’t yet discussed on AAS Nova from among the most-downloaded papers published in AAS journals this year. The usual posting schedule will resume in January.

Identifying Exoplanets with Deep Learning: A Five-Planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90

Published January 2018

Main takeaway:

Using machine learning, Google Brain engineer Christopher Shallue and NASA Sagan Postdoctoral Fellow Andrew Vanderburg (The University of Texas at Austin, Harvard-Smithsonian CfA) have discovered two new planets within previously known Kepler multi-planet systems.

Why it’s interesting:

training light curves

Examples of light curves used to train the authors’ neural network models. [Shallue & Vanderburg 2018]

In today’s field of exoplanet astronomy, observatories like Kepler and TESS have guaranteed us plenty of data. But the transit signal of an Earth-sized planet around a Sun-like star remains at the edge of detectability, and our best bet for reliably picking such signals out of the noise is automation. Shallue and Vanderburg’s study demonstrates the power of training a deep convolutional neural network to identify planet signals in data like Kepler’s.

What was found:

Shalle and Vanderburg’s models identified two signals from among data from previously known Kepler multi-planet systems: one planet that is part of a five-planet resonant chain around Kepler-80, and one planet orbiting Kepler-90. Kepler-90 was previously known to host seven planets, so this discovery of an eighth has brought Kepler-90 to a tie with our own Sun for the star known to host the largest number of planets.

Citation

Christopher J. Shallue and Andrew Vanderburg 2018 AJ 155 94. doi:10.3847/1538-3881/aa9e09