Google and university researchers use in-depth learning to discover exoplanets



[ad_1]

Researchers from Google and several universities have discovered two new exoplanets (planets outside our solar system) through a convolutional neural network called AstroNet K2. 14 additional objects could also be identified as exoplanets with additional searches.

This announcement builds on research published last year by astrophysicist Andrew Vanderburg of Harvard University and Chris Shallue of Google AI, who also uses machine learning to screen NASA's Kepler data and find the celestial bodies in space. Later, Google opened its model for searching for exoplanets using Kepler data on GitHub.

"The work is important because it's the first time that a neural network has been successfully applied to K2 data," said Anne Dattilo, research assistant at the University of Texas, when of a telephone interview. "Different types of machine learning have been applied to different types of astronomical datasets, as its predecessor was based on Kepler data, but K2 data presents different challenges because the telescope was unstable."

For the first four years after its launch in 2009, the Kepler Space Telescope was used to study potential Earth-like planets that pass in front of the stars. The telescope observed more than 200,000 stars, but a mechanical malfunction prevented it from focusing on only one part of the sky, which resulted in more sporadic data collection. Kepler was officially withdrawn by NASA last year.

To meet this challenge, more than 30,000 images with promising features were collected and reviewed, and more than 22,000 images were used to form the semi-supervised AI system. AstroNet K2 is 98% accurate in test data sets.

Members of the Google Brain team; the astronomy departments of the University of California at Berkeley and the University of Texas at Austin; and the Harvard-Smithsonian Center for Astrophysics presented the results in a paper. They conclude that AstroNet K2 is "not yet ready to automatically detect and identify candidates from the planet" because of the identification of too many false positives, but this could increase the efforts of astronomers working to better understand the universe.

"It's not enough to give us a handful of candidates and say," It's them. They are planets, and that's all. It returns a whole lot, mitigated by false positive signals. So you need the help of a human astronomer to sort out this information and see what is not a planet. But instead of 20,000 signals, you only have to examine 1,000 signals, which saves a lot of time, "she said.

Like its predecessor, AstroNet K2 will be refined and opened to be made available to the AI ​​community in the future, Dattilo said.

[ad_2]

Source link