A team of astronomers from the University of Hawaiʻi at Mānoa’s Institute for Astronomy (IfA) has produced the most comprehensive astronomical imaging catalog of stars, galaxies, and quasars ever created with help from an artificially intelligent neural network.
The group of astronomers from the University of Hawaiʻi at Mānoa’s Institute for Astronomy (IfA) released a catalog containing 3 billion celestial objects in 2016, including stars, galaxies, and quasars (the active cores of supermassive black holes). Needless to say, the parsing of this extensive database—packed with 2 petabytes of data—was a task unfit for puny humans, and even grad students. A major goal coming out of the 2016 catalog release was to better characterize these distant specks of light, and to also map the arrangement of galaxies in all three dimensions. The Pan-STARRS team can now check these items off their to-do list, owing to the powers of machine learning. The results of their work have been published to the Monthly Notices of the Royal Astronomical Society.
Their PS1 telescope, located on the summit of Haleakalā on Hawaii’s Maui island, is capable of scanning 75% of the sky, and it currently hosts the world’s largest deep multicolor optical survey, according to a press release put out by the University of Hawaiʻi. By contrast, the Sloan Digital Sky Survey (SDSS) covers just 25% of the sky.
To provide the computer with a frame of reference, and to teach it how to discern celestial classes of objects from one another, the team turned to publicly available spectroscopic measurements. These measures of colors and sizes of objects numbered in the millions, as Robert Beck, the lead author of the study and a former cosmology postdoctoral fellow at IfA, explained in the press release.
“Utilizing a state-of-the-art optimization algorithm, we leveraged the spectroscopic training set of almost 4 million light sources to teach the neural network to predict source types and galaxy distances, while at the same time correcting for light extinction by dust in the Milky Way,” Beck said.
These training sessions worked well; the ensuing neural network did a bang up job when tasked with sorting the objects, achieving success rates of 98.1% for galaxies, 97.8% for stars, and 96.6% for quasars. The system also determined the distances to galaxies, which were at most only off by about 3%. The resulting work is “the world’s largest three-dimensional astronomical imaging catalog of stars, galaxies and quasars,” according to the University of Hawai’i.
“This beautiful map of the universe provides one example of how the power of the Pan-STARRS big data set can be multiplied with artificial intelligence techniques and complementary observations,” explained team member and study co-author Kenneth Chambers. “As Pan-STARRS collects more and more data, we will use machine learning to extract even more information about near-Earth objects, our Solar System, our Galaxy and our Universe.”
The new catalog, which was made possible by a grant from the National Science Foundation, is publicly available through the Mikulski Archive for Space Telescopes. The database is 300 gigabytes in size, and it’s accessible through multiple formats, including downloadable computer-readable tables.
This survey has already yielded some interesting science, including an explanation for a rather spooky region of space known as the Cold Spot. Using the PS1 telescope, and also NASA’s Wide Field Survey Explorer satellite, the Pan-STARRS scientists spotted a massive supervoid—a “vast region 1.8 billion light-years across, in which the density of galaxies is much lower than usual in the known universe,” as the University of Hawai’i described it five years ago. It’s this supervoid that is causing the Cold Spot, as it’s seen in the cosmic microwave background, according to the researchers.
The updated map will also be used to study the overall geometry of the universe, to further test our theories about the standard cosmological model, and to analyze ancient galaxies, among many other avenues of astronomical and cosmological research.