When galaxies collide with each other it's a very slow process, which can take billions of years to complete. It is crucial for the evolution of galaxies but what we know comes from looking at different objects. We have to look at different snapshots to try to build a full picture of what's going on.
A paper recently published in Astronomy & Astrophysics used artificial intelligence (AI) to study mergers from 12 billion years ago until today. A total of 200,000 mergers, a truly record-breaking number, were analyzed and the researchers found that while merging galaxies don't have particularly high star-formation rates, they do lead to increased starburst phases.
Starbursts are localized episodes of intense star formation and in the team's sample, there were twice as many starbursts in merging galaxies compared to in single galaxies. The researchers stress how crucial the introduction of neural network algorithms has been to achieve their results. This marks the first time the approach has been used to study galaxy mergers.
"This is a milestone in the sense that AI will play an increasingly large role in our field," team leader Dr Lingyu Wang, from SRON Netherlands Institute for Space Research, said in a statement. "But we have to keep in mind that the power of AI is limited to how it is trained. If we feed it a flawed definition of a galaxy merger, then it won't do its job correctly."
The limitations of AI have been a significant issue in implementing these approaches. For many years, humans were much better at spotting the hallmarks of galaxy mergers than automated approaches. Now, AI seems to have caught up.
To deal with the huge database, the team had do build a machine-learning algorithm that taught itself what merging galaxies looked like. It was good enough to deliver a very simple but effective binary classification of merger/non-merger.
"The advantage of artificial intelligence is that it improves the reproducibility of our study because the algorithm is consistent in its definitions of a merger. Also, it's good preparation for upcoming surveys that will image billions of galaxies," first author William Pearson explained. "Then you inevitably need AI. Even citizen science projects such as Galaxy Zoo cannot deal with those numbers."
2