Speaker
Description
We present a transformer-based autoregressive foundation model specifically designed to analyze narrow-band photometry from the Javalambre Physics of the Accelerating Universe Astrophysical Survey (J-PAS), simulated using 18 million spectra from the Dark Energy Spectroscopic Instrument (DESI). By crossmatching DESI value-added catalogs, our model incorporates detailed stellar population properties, including stellar masses and star formation rates (SFR) for galaxies, emission line fluxes, and photometric redshifts for both galaxies and QSOs. Additionally, the model estimates fundamental stellar parameters such as effective temperature, surface gravity, and alpha enhancement. A significant advantage of our approach is its simultaneous capability for classification and regression tasks, allowing the model to classify astronomical objects while accurately predicting their physical parameters. This unified method eliminates the need for multiple specialized codes and significan
tly improves inference speed compared to traditional techniques. The transformer's architecture also enhances interpretability by explicitly modeling the relationship between physical properties and galaxy, QSO, and stellar spectra. Its autoregressive nature efficiently handles incomplete or partially missing data, facilitating straightforward integration of multi-wavelength datasets. Furthermore, training the model with simulated J-PAS fluxes from DESI spectra alongside real observational data inherently supports effective domain adaptation.
Beyond standard predictive tasks, our model enables broader applications, including identifying rare astronomical objects. It can also perform similarity searches, identifying objects with comparable spectral and physical properties, enhancing its versatility for various scientific analyses within J-PAS and future data-intensive astronomical surveys. Consequently, a comprehensive value-added catalog covering galaxies, QSOs, and stars will be publicly released, offering an essential resource for diverse astrophysical studies.
