31 August 2026 to 4 September 2026
University of Malta
Europe/Malta timezone

Needles in the cosmic haystack: Innovating the search for Reionization-Era quasars with self-supervised machine learning

2 Sept 2026, 11:30
20m
Aula Prima (University of Malta)

Aula Prima

University of Malta

Valletta Campus, St Paul Street Valletta VLT 1216, Malta

Speaker

Laura Natalia Martínez Ramírez

Description

Luminous quasars at the highest redshifts (z>6) are key laboratories for studying early supermassive black hole (SMBH) growth and the physical conditions of the Universe during cosmic reionization. However, their extremely low spatial density (< 1 per Gpc³), combined with severe contamination from foreground ultracool dwarfs that outnumber them by up to four orders of magnitude, makes their discovery as challenging as finding a needle in a haystack. In this work, we leverage the extensive coverage of Legacy Survey DR10 to conduct a pioneering self-supervised search for quasars at z > 6, directly analyzing multi-band optical images, while minimizing the biases of traditional catalog-driven color selections. Using a contrastive learning approach, we successfully identify an overdensity of quasars at z ∼ 6.3 within the latent space of i-dropout candidates, achieving an almost 1:1 contamination ratio with brown dwarfs. In the first spectroscopic campaign, we have already confirmed 16 new quasars at 5.94 < z < 6.45, some of which exhibit narrow Lyman-alpha emission lines (FWHM ~400-1100 Km/s), suggesting potential high dust obscuration or early SMBH growth stages. These results highlight the effectiveness of self-supervised machine learning techniques, combined with SED-fitting-based prioritization, in uncovering rare and distant astrophysical sources beyond the limitations of conventional methods. Our approach establishes a scalable framework to advance the census of high-redshift quasars and SMBH demographics, with demonstrated applications to 4MOST-ChANGES campaign and Euclid data, and clear potential for forthcoming surveys such as Rubin/LSST and Roman, ultimately improving constraints on SMBH formation and evolution within the first Gyr of the Universe.

Presentation materials