Using a new deep learning algorithm and data from the Galaxy Zoo citizen science project, astronomers have discovered 40,000 ring galaxies.
“Galaxies live a chaotic life,” said lead author Dr. Mike Walmsley, a postdoctoral researcher at the University of Manchester, applying deep learning research breakthroughs to astrophysics.
“Collisions with other galaxies and bursts of energy from supermassive black holes disrupt the colors and orbits of billions of stars, leaving tell-tale markers that volunteers search for on the Galaxy Zoo website.”
“But understanding exactly which cosmic events lead to which markers requires millions of measured images — more than humans could ever search.”
For their new research, Dr. Walmsley and colleagues used a decade of Galaxy Zoo measurements.
“Galaxy Zoo turns 15 years old this week, and we are still innovating,” said Galaxy Zoo deputy principal investigator Dr. Brooke Simmons, an astronomer with the University of Lancaster.
“This work will make it possible for a new generation of discoveries to be made from upcoming large galaxy surveys.”
Images of 35 of the ring galaxies discovered by the Galaxy Zoo volunteers and the Zoobot. Image credit: Mike Walmsley / Galaxy Zoo Collaboration.
The astronomers created a new deep learning algorithm — named Zoobot — that can not only accurately predict what Galaxy Zoo volunteers would say but understands where it might be mistaken.
“Zoobot is designed to be retrained again and again for new science goals,” they said.
“Just like a musician can learn a new instrument faster than their first instrument, Zoobot can learn to answer new shape questions easily because it has already learned to answer more than 50 different questions.”
“With Zoobot, humans and machines are collaborating to push the science of astronomy forward,” Dr. Walmsley said.
“We’re helping other astronomers solve questions we never thought to ask.”
As a results, the team discovered 40,000 rare ring-shaped galaxies — six times more than previously known.
“Rings take billions of years to form and are destroyed in galaxy-galaxy collisions, and so this giant new sample will help reveal how isolated galaxies evolve,” the authors said.
“The dataset will also tell scientists how galaxies age more generally.”