For the last 12 years, the “Galaxy Zoo” has been working hard to improve our understanding of the cosmos. Despite its name, the Galaxy Zoo doesn’t house any aliens; it is, instead, a crowdsourced astronomy project that asks its users to classify the shapes of massive numbers of galaxies. Now, researchers from the National Center for Supercomputing Applications (NCSA) and the Argonne Leadership Computing Facility (ALCF) are using AI and supercomputing to leverage that user-generated data and accelerate progress on the Galaxy Zoo.

The research team developed a new approach to classifying these hundreds of millions of galaxies. Instead of relying on crowdsourced classification, the researchers used knowledge from the state-of-the-art Xception neural network, combined with the datasets generated by the Galaxy Zoo project, to train its deep learning models. They then applied the trained model to galactic images from the Dark Energy Survey (DES) – where it achieved a 99.6% accuracy in identifying spiral and elliptical galaxies.

“Using the millions of classifications carried out by the public in the Galaxy Zoo project to train a neural network is an inspiring use of the citizens science program,” said Elise Jennings, a computer scientist at ALCF. “This exciting research also sheds light on the inner workings of the neural network, which clearly learns two distinct feature clusters to identify spiral and elliptical galaxies.”

The researchers extracted the overlapping images from the two datasets using the NCSA’s Blue Waters supercomputer, then taught their deep learning model on the Pittsburgh Supercomputing Center’s Bridges supercomputer. The team also used the K80 Nvidia GPUs in the Cooley supercomputer at ALCF to reduce the training stage for the Xception model from five hours to eight minutes.

“We’re excited to work with the team at NCSA and Argonne as well as the researchers who drove the original Galaxy Zoo effort to pursue this important area of scientific discovery,” said Tom Gibbs, manager of developer relations at Nvidia. “Using these new methods, we’re taking an important step to understanding the mystery of dark energy.”

About the research

The study referenced in this article, “Deep learning at scale for the construction of galaxy catalogs in the Dark Energy Survey,” was written by Asad Khan, E.A. Huerta, Sibo Wang, Robert Gruendl, Elise Jennings and Huihuo Zheng. It was published in the August 2019 issue of Physics Letters B and is accessible online at this link.