Machine learning finds 300 massive galaxy clusters hiding in microwave maps

Researchers applied a convolutional neural network to combined microwave maps from the Planck and ACT telescopes to identify galaxy clusters via their distinctive thermal signatures. They confirmed roughly 60% of the 2,962 candidates as real clusters and measured their masses and distances using standard scaling relations; the sample reaches to high redshift and finds ~300 ultramassive systems (over 10^15 solar masses) at z>0.7, roughly 10% more than previously catalogued. This demonstrates that deep learning can squeeze additional discoveries from existing data, extending our inventory of the universe's most extreme gravity laboratories.