Speaker
Description
We present a new value-added parameter catalog for the LAMOST (Large Sky Area Multi-Object Fiber Spectroscopic Telescope) survey, produced by a spectral foundation model that unifies low- and medium-resolution LAMOST spectra with high-precision labels from multiple high-resolution surveys. The model, built upon the SpecCLIP framework, learns a shared latent space among spectra of different resolutions and predicts stellar atmospheric parameters and elemental abundances with high accuracy. Trained with more than one million cross-matched spectra from LAMOST, APOGEE, GALAH, Gaia-ESO, and H3 surveys, the model achieves consistent parameter estimation across resolutions and significantly improves the precision of effective temperature, surface gravity, metallicity, and α-element abundance. The first release of this AI-driven parameter catalog has been integrated into the LAMOST data release pipeline and will be continuously updated in future data releases. This work demonstrates how large-model-driven spectral analysis can enhance traditional survey pipelines and open new opportunities for automated discovery in large-scale stellar spectroscopy.
| Affiliation of the submitter | National Astronomical Observatories, Chinese Academy of Sciences |
|---|---|
| Attendance | in-person |
