Finding challenging worlds and discovering many other kinds of dips and peaks in astronomy data is becoming more and more the purview of software instead of humans. Feed a machine learning algorithm a few tens of thousands of labeled galaxy images, and it will happily catalog another 27 million galaxies. Seriously, this is a thing that happened, and the work, led by Jesús Vega-Ferrero and Helena Domínguez Sánchez and published in Monthly Notices of the Royal Astronomical Society (MNRAS), is giving us an amazing census of the kinds of galaxies that have occupied the universe in the past six billion years. 

While classifying galaxies is time-consuming, this problem isn’t actually hard for humans. Where machine learning algorithms are really excelling is in searching through massive datasets for extremely rare objects. Like playing a celestial game of Where’s Waldo, it is often easy to look right at the kind of thing you’re looking for and not see it among the stars. 

More Information

University of Pennsylvania press release

“Pushing automated morphological classifications to their limits with the Dark Energy Survey,” J Vega-Ferrero et al., 2021 March 2, Monthly Notices of the Royal Astronomical Society