Kepler’s New Findings
Today, NASA announced the latest discovery made using its Kepler space observatory. They found a system of eight planets, proving once and for all that solar systems like our own exist in the cosmos. During the press conference, the agency emphasized that the find was only made possible because of the deep learning techniques developed by Google. These techniques allowed them to analyze the data the spacecraft has been collecting and, ultimately, revealed the new find.
In attendance at the briefing was the director of NASA’s astrophysics division, Paul Hertz, and Kepler project scientist Jessie Dotson. Also present were NASA Sagan postdoctoral fellow Andrew Vanderburg and Google AI engineer Christopher Shallue.
This study came from work instigated by Shallue during his tenure as a senior software engineer at Google Brain. He posed the question of whether deep learning could be used to discover new planets, and went on to collaborate with Vanderburg in order to find out, producing convolutional neural networks that could identify exoplanets in data collected by Kepler.
Vanderburg had been working on extrasolar planet detection and characterization during his Ph.D. study at Harvard University. He developed analysis techniques for photometry data recorded by Kepler, pursuing ground-based observations of interesting planets and binary stars.
Sifting Through the Data
Between its initial mission and the initiative dubbed ‘Second Light’ that came about as a result of equipment failure, Kepler has collected a huge amount of data. By 2016, it had confirmed the existence of 1,284 new planets – but far more candidates had been established, with a total of 4,696 being listed in a planet candidate catalog released in July 2015.
It takes a lot of time and effort to pore through this data by hand. Leveraging machine learning to take care of the grunt work means that human researchers can spend more of their time working on projects that simply can’t be automated.
NASA’s budget isn’t unlimited, and this kind of implementation of machine learning could help it go further. The money the agency saves might not be enough to pay for a mission to Mars, but these extra resources can be applied to other research.
Much has been said about the capacity for Ai and machine learning to swallow up jobs that were once performed by humans. However, this kind of implementation of such technology demonstrates how it can carry out certain tasks far more efficiently, and potentially with greater accuracy.
Kepler has offered us a great deal of information about distant planets and exoplanets. While we’re yet to find concrete evidence that these worlds can support life, we’re closer to being able to make that kind of deduction thanks to the spacecraft.