Program: GN-2024B-Q-215

The search for high redshift quasars over the past two decades has led to major discoveries in areas from supermassive black hole accretion to cosmic reionization. The significance of these scientific results has relied on the completeness and reliability of the quasar selection method from optical-to-NIR broad-band colors. Recently, machine learning has been forecasting improvement in both domains for the color selection of high-redshift quasars. Utilizing multi-color photometry, computers are trained to tell quasars from stars down to a regime previously missed by simple color cuts. In our previous work, we identified 35 candidates of quasars at z~5 down to M1450=-22 mag, using machine learning and Bayesian information criterion. The number of candidates is nearly 2x than what another group identified using a traditional color box selection method. Through Gemini/GMOS observations, we aim to constrain the selection efficiency of faint z~5 and 6 quasars, based on the combination of machine learning and Bayesian information criterion. We obtained the Gemini/GMOS time in 23B and 24A semesters to observe 7 targets, but only one target has been observed so far. Here, we propose to observe 5 faint (M1450 < - 22.1 mag) quasar candidates observable in 24B that cannot be selected with the simple color box selection. The combination of the targets from 23B,24A and 24B enables a statistically meaningful test of the selection method at the fainter magnitudes, resulting in a reduction of the standard error by 23.5%. If the quasar selection efficiency turns out to be high at less explored color or luminosity space, it will validate and promote the use of machine learning in finding distant quasars down to the survey limit. The confirmed quasars will better constrain the faint-end luminosity function by bridging the gap between the bright quasar LF from wide-field surveys and recent JWST results.