A brainy look at dark matter
 |
|
| Galaxy shape reconstruction algorithm. From left to right: the original image including atmospheric distortion and noise fluctuations, reconstruction of the image with noise removed, and reconstruction with atmospheric effects and pixelation removed. Images courtesy of David Kirkby |
|
Have you ever sat in an open field at night, looked up at the vast number of stars and thought, “I bet an artificial brain would come in handy for making sense of all this”? You might if you were planning the best way for NASA to map the sky.
University of California, Irvine professor David Kirkby and his doctoral student Daniel Margala tapped an analytical technique from particle physics to win a contest to help NASA's Euclid mission map galaxy clusters.
The technique? An artificial neural network like the ones Kirkby used in his work on the BaBar experiment at SLAC National Accelerator Laboratory. It's just one of many connections he has been finding between his past collider work and current astrophysics projects, including the Large Synoptic Survey Telescope and BOSS, the Baryon Oscillation Spectroscopic Survey, which is mapping galaxies and quasars to get a better handle on the expansion of the universe.
Where once he sought answers to nature's fundamental questions in the debris from particle collisions, “Now I'm tackling the same questions using telescopes,” Kirkby says. For someone breaking into a new field, he adds, the contest offered an “ideal opportunity.”
In creating the contest, NASA's Jet Propulsion Laboratory and the United Kingdom's Royal Astronomical Society teamed up with Kaggle, a company that hosts competitions to solve statistical and analytical problems. JPL needed an innovative system for mapping dark matter in upcoming missions such as the Euclid telescope survey, which will map the large-scale structure of the universe by imaging an unprecedented number of faint galaxies whose shapes are subtly distorted by the presence of dark matter. One of the grand challenges in cosmology is to analyze these distortions precisely enough to unravel the interplay between dark matter and dark energy, the mysterious force that is thought to fuel the accelerated expansion of the universe.
Kaggle provided contestants with 100,000 images that were intentionally blurred and contained various amounts of noise. Kirkby searched his particle physics background for approaches he could use. He and Margala submitted a total of 16 data sets, gradually improving their score and beating out the other 72 teams.
The prize? A trip to JPL to present their results. But the ultimate reward may come in 2019, when Euclid is scheduled to launch.
Brad Hooker
Click here to download the pdf version of this article.
