Spectra confirm 70 gravitational lenses that a neural network flagged in DESI Legacy Surveys images

The discovery came in two steps ^[1]. First, a residual neural network searched the survey data for the signatures of lensing ^[8]. It is a type of machine-learning model designed to recognise subtle patterns in images ^[8]. The pattern it looks for is an arc, a ring or a repeated image, which forms when a foreground galaxy or cluster bends and magnifies light from something behind it ^[12]. The team used it because thousands of lenses were likely hiding in the data and searching for them by eye was impractical ^[14].

The second step was spectroscopy. A follow-up observing program led by Aleksandar Cikota, an associate scientist at NSF NOIRLab, took the spectra that established 70 candidates as true lenses ^[11][2].

Earlier work with the same techniques had built a catalogue of about 3,500 potential lenses ^[9]. Seventy confirmations are about 2 percent of that list ^[15]. That figure measures progress through a queue. It is not a purity score, because a candidate that has not yet had a spectrum is neither confirmed nor rejected. Phys.org reports that the machine-learning search let astronomers focus telescope time on follow-up observations ^[16].

The August map is a separate question. Released in August 2026, it holds nearly 4 billion stars, galaxies, black holes and asteroids across 5.6 trillion pixels ^[4]. It combines the DECaLS, MzLS and BASS surveys, with years of data from NASA's WISE satellite added ^[5]. Phys.org describes the map as 13 years in the making, with regular releases along the way, and introduces the lens search as one of many projects those releases made possible ^[6]. The report does not say which release the candidates came from ^[6]. I think the credit for these 70 lenses belongs to the survey's long record of releases. This paper does not show how quickly the August map will produce more.

The confirmed lenses have real uses. Lenses can magnify galaxies from the early universe, reveal how invisible dark matter is distributed, improve measurements of the universe's expansion rate and help detect exoplanets ^[13]. "What we're building now is a carefully confirmed sample that researchers can use for years to come," Cikota said ^[10]. The 70 join what phys.org calls one of the largest collections of confirmed lenses to date ^[3].