Wednesday 09 April 2025
A team of astronomers has made significant progress in their search for galaxy-scale lenses, a phenomenon where the gravity of a foreground object warps the light from a background source, creating multiple images or even magnifying distant galaxies. Using data from the Hyper Suprime-Cam (HSC) survey and advanced machine learning techniques, they’ve identified 95 new lens candidates with high confidence.
The search for galaxy-scale lenses is crucial for understanding the nature of dark matter and dark energy, which are thought to make up about 95% of the universe’s mass-energy budget. By studying these lenses, astronomers can gain insight into the distribution of matter in the universe and how it has evolved over time.
To identify lens candidates, the team used a combination of machine learning algorithms and visual inspection. First, they trained five neural networks to classify images from the HSC survey as potential lenses or non-lenses. The networks were fed a dataset of known lenses and non-lenses, allowing them to learn patterns in the data that distinguish between the two.
The trained networks then classified over 110 million images from the HSC survey, producing a list of candidates with high confidence scores. However, this initial list included many false positives, which needed to be removed before further analysis could begin.
To address this issue, the team implemented several techniques to clean up the candidate list. They used SExtractor, a software package that identifies sources in astronomical images, to reject objects that were likely not lenses. They also employed a modeling network from their previous work to refine the classification of remaining candidates.
After these steps, 3,408 lens candidates remained. The team then performed a comprehensive visual inspection of these candidates, involving eight individuals who carefully examined each image and assigned a grade based on its likelihood of being a genuine lens.
The final list of lens candidates included 95 systems with high confidence scores, including 92 discoveries that are reported for the first time. These lenses can be used to study the properties of dark matter and dark energy, as well as the distribution of normal matter in the universe.
The success of this project demonstrates the power of combining machine learning techniques with human expertise in astronomical research. By automating the initial classification step and using advanced algorithms to refine the candidate list, the team was able to efficiently identify a large number of lens candidates that would have been difficult or impossible to find through traditional methods alone.
Cite this article: “Unlocking the Secrets of the Cosmos: A New Lens on the Universe Revealed by HOLISMOKES XVI”, The Science Archive, 2025.
Astronomy, Galaxy-Scale Lenses, Machine Learning, Dark Matter, Dark Energy, Hyper Suprime-Cam, Hsc Survey, Neural Networks, Sextractor, Gravitational Lensing
