With telescopes transmitting an entire universe of images back to Earth, human brains can only take so much — which is where an artificial brain comes in.
Astronomers may soon be getting help sorting through all that data from a computerized “deep learning” algorithm, a method that has previously been used for face and speech recognition. Something that can recognize your face has the potential to recognize bodies of stars, which is exactly what Joel Primack, professor emeritus of physics at UC Santa Cruz, and colleagues were thinking when they used the system to recognize three phases of galaxy evolution. It was scary accurate.
"We were not expecting it to be all that successful. I'm amazed at how powerful this is," said Primack, whose study was recently published in Astrophysical Journal. "We know the simulations have limitations, so we don't want to make too strong a claim. But we don't think this is just a lucky fluke."
Primack and his team trained the algorithm by using output from digital simulations of galaxy formation, creating images similar to phenomena Hubble has observed. The system learned to recognize the pre-blue nugget phase, the blue nugget phase, and the post-blue nugget phase before it was challenged with actual Huggle images.
Galaxies show themselves as they were millions and billions of years ago depending on how distant they are. What phase of galactic evolution we see depends on how long it took the starlight from a particular galaxy to travel through the vastness of space, which is why astronomers visually travel back in time as a space telescope ventures deeper and deeper into the unknown. Simulation is the only way to follow that evolution (short of immortality) and find out how a galaxy came into being before Earth even knew what a dinosaur was.
The blue nugget phase that appeared in Primack’s simulations is thought to have occurred in the early lives of galaxies that had enough gas to form a dense star-spawning region. This phase indicates active star formation because of the “blue” wavelengths of light emitted by blazing young stars, and is only found in galaxies with a particular mass range. The algorithm’s ability to consistently identify that by itself suggests that it is seeing a physical process that really is happening out there.
"We wanted to pick a process that theorists can define clearly based on the simulations, and that has something to do with how a galaxy looks, then have the deep learning algorithm look for it in the observations,” said study coauthor and fellow UC Santa Cruz professor emeritus David Koo.
While this is a breakthrough, the simulations still can’t tell the algorithm. Active galactic nuclei — the ultra-luminous centers of galaxies where supermassive black holes devour gas — still need to be investigated because they are believed to regulate star formation.
“We're just beginning to explore this new way of doing research,” said Koo. “It's a new way of melding theory and observations."
(via UC Santa Cruz)