Artificial intelligence (AI) is being applied to a number of fields, but just recently, a group of researchers managed to train a deep learning algorithm — a branch of AI — to analyze images of distant galaxies and reveal how they formed and evolved over time.

Understanding galactic evolution is one of the key puzzles in gaining more insight into the formation of our universe. We have a bunch of ground and space-based telescopes that can peer through the cosmos and capture these galaxies, but understanding every stage of evolution for an individual galactic candidate hasn’t entirely been possible.

This is because galaxies change their face over several billion years and our telescopes can only show how a galaxy looked at one particular period of time. As light from distant space objects takes millions to billions of years to travel, we always have the option to peer deeper into the cosmos and look back in time at other younger galaxies. But, for one specific target, the only thing currently possible is to capture images and then compare them with computer-based simulations to predict how they might have changed.

This is where deep-learning algorithm comes in. Scientists from University of California Santa Cruz recently fed the program with images of computer simulated galaxies and trained it to recognize three crucial stages of evolution. On being tested on actual images from Hubble Space Telescope, the program identified its own patterns and delivered very accurate results.

"We were not expecting it to be all that successful. I'm amazed at how powerful this is," co-author Joel Primack said in a statement. "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."

During the study, the group focused on a phenomenon dubbed blue nugget — a dense, star-forming region where hot stellar bodies emit light in blue wavelengths of light. The process has been seen in several simulations of young gas-rich galaxies, but the algorithm not only classified the galaxy but also defined its own pattern — in both simulated and observed data — and noted the process occurs in galaxies within a certain range of mass. It even figured that following blue nugget, a compact red nugget phase starts because star-formation slows down in the galaxy and only cooler bodies are left behind.

"It may be that in a certain size range, galaxies have just the right mass for this physical process to occur," co-author of the work David Koo said in the same statement.

 "This project was just one of several ideas we had," Koo added. "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. We're just beginning to explore this new way of doing research. It's a new way of melding theory and observations."

That said, the algorithm can reveal complex features in observed datasets, things that we humans cannot see or understand. However, there is still no way to understand the basis of these revelations, just like in this case. More studies will be needed in the future to further develop and train the system.

The paper, titled “Deep Learning Identifies High-z Galaxies in a Central Blue Nugget Phase in a Characteristic Mass Range,” is set to be published in the Astrophysical Journal.