Sometimes, when a job is overwhelming, it takes a machine to do it well. That's why NASA recently turned to computers and machine learning to gather data about the thousands of stars in our galaxy.
As our technology grows and our telescopes see farther, we now have huge sets of data on stars, but not enough manpower to go through all that data. Even worse, determining basic properties of stars, such as size and composition, means studying the light from stars at different wavelengths. That is an arduous process and requires a lot of hours to complete. It also requires expensive equipment.
However, using a computer to sort through all that data is much faster, and the computer can look for patterns that identify certain information about a star. So basically, using a special algorithm and machine learning, a computer can give us details on billions of stars in a lot less time and with a lot less money.
NASA compares this sort of machine learning to what Netflix does when it recommends movies and TV shows to its individual users.
"It's like video-streaming services not only predicting what you would like to watch in the future, but also your current age, based on your viewing preferences," says Adam Miller of NASA's Jet Propulsion Laboratory. "We are predicting fundamental properties of the stars."
Of course, this won't work if the machine doesn't know what it's looking for. And that means that humans must teach machines the patterns specific to certain kinds of stars. So NASA researchers spent time plugging in data for about 9,000 stars that we already had certain kinds of information on, including their sizes, temperatures and compositions. In addition to that, researchers also input data about the "varying brightness" of those stars, recorded by the Sloan Digital Sky Survey. This additional information shows the computer the stars' light curves. By having both sets of data, the machine learns how to pair certain light curves with certain star properties.
After that, a computer algorithm allows the machine to start making its own predictions by just looking at stars' light curves. Additional input from humans makes the machine's predictions even more accurate, until the machine is able to give information about each star on its own.
This is the first time researches have applied machine learning for determining specific properties of specific stars by looking at photos of those stars taken at different periods in time. This gives us new insight into learning about each star's birth, as well as its evolution to the celestial body it is now.
"With more information about the different kinds of stars in our Milky Way galaxy, we can better map the galaxy's structure and history," says Miller.
[Photo Credit: NASA, ESA, Hubble, HPOW]
